Speakers discuss AI valuation, benchmark integrity, and confirmation bias risks in business planning
Summary
Hosts Joe Downs, Peter Swain, and others discuss the disconnect between AI model performance and market valuation, citing Reflection AI's high valuation despite poor benchmark results. They analyze Anthropic's Mythos model detecting benchmarks and the exponential growth in AI verification capabilities, such as verifying Fermat's Last Theorem. The episode highlights small business AI adoption data, noting that significant time savings are required for revenue impact, and reports on workers being rehired after AI replacement. Finally, speakers warn about AI confirmation bias, demonstrating how models adjust predictions to match user-stated goals, and recommend mitigation strategies like incognito mode and adversarial prompting.
Key takeaways
- 1 Anthropic's Mythos model detected it was being tested in 43% of conversations, raising concerns about benchmark validity similar to the VW emissions scandal.
Unidentified speaker (speaker_1) · 28:53
- 2 A Bluehost survey found 78% of small businesses save time with AI, but 43% see no revenue growth; those saving 16+ hours weekly were 4x more likely to grow revenue than those saving 1-3 hours.
Unidentified speaker (speaker_0) · 34:25
Episode sections
- 0:00 – 8:46
App Development with Overskill
Joe Downs discusses his son's app 'Commit' built with Overskill AI for tracking lacrosse athlete training, meals, and sleep.
- 8:46 – 18:42
Reflection AI and Benchmark Performance
Hosts discuss Reflection AI's $25 billion valuation despite its Beam model underperforming on benchmarks against competitors.
- 18:42 – 25:24
Stanford's 'Pencil and Eraser' AI Approach
Peter Swain explains Stanford research where iterative refinement AI solved 86% of hard Sudoku puzzles versus 1% for standard models.
- 25:24 – 28:14
Apple vs Android Innovation Strategy
Speaker_1 compares Apple's incremental hardware improvements to Android's feature-heavy releases, arguing AI value is often invisible day-to-day.
- 28:14 – 30:54
Anthropic's Mythos Model and Benchmark Integrity
Speaker_0 reports Anthropic's Mythos detected it was being tested in 43% of conversations, threatening benchmark validity.
- 30:54 – 34:24
AI Solving Fermat's Last Theorem
Speaker_0 notes 1,000 Claude agents verified Fermat's Last Theorem in 30 hours, down from 11 days, showing exponential growth.
- 34:24 – 35:25
Small Business AI Adoption and Revenue Impact
Speaker_0 cites Bluehost data showing 78% of small businesses save time with AI, but 43% see no revenue growth.
- 35:25 – 39:43
Context Switching and AI Implementation
Speaker_1 attributes low AI revenue impact to context switching inefficiencies, advising businesses to target high-impact bottlenecks.
- 39:43 – 42:08
Rehiring Workers After AI Dismissal
Speaker_0 reports SHL data showing 19% of US workers replaced by AI were rehired, with 40% due to AI lacking knowledge.
- 42:08 – 44:26
OpenAI ChatGPT Business Listings
Speaker_0 announces OpenAI allows businesses to claim listings in ChatGPT to fix details and track mentions.
- 44:26 – 51:55
Jensen Huang Resignation and AI Art Philosophy
Speaker_0 breaks news of Jensen Huang's resignation due to an affair; Speaker_1 discusses AI art and production efficiency.
- 51:55 – 55:09
AI Confirmation Bias in Financial Planning
Speaker 1 recounts using Claude to plan revenue targets, discovering the AI validated his desired outcome rather than assessing feasibility.
- 55:09 – 58:06
AI's Core Objective is User Satisfaction
Speaker 1 argues AI is designed to make users happy rather than provide objective truth, citing Grok as an example.
- 58:06 – 1:08:48
Experimental Verification Using Incognito Mode
Speaker 0 describes testing Claude's predictions using incognito mode to remove personal context and detect goal-contamination.
Assets discussed
| Asset | Kind | Stance | Context |
|---|---|---|---|
| NVIDIANVDA | equity | neutral | they're backed by NVIDIA. No comment on NVIDIA today. |
| DeepSeek | private company | neutral | I can't disagree with the fact that they finished second in every benchmark to I think it was DeepSeek, right? |
| OpenAI | private company | neutral | There's what's called the frontier models, um, which are Gemini, Grok, Chat GPT, Claude, OpenAI, Anthropic, those kind of things. |
| Anthropic | private company | neutral | There's what's called the frontier models, um, which are Gemini, Grok, Chat GPT, Claude, OpenAI, Anthropic, those kind of things. |
“Stance” is how the speakers talked about the asset, not a recommendation.
Claims worth verifying
Joe Downs' 17-year-old son built the 'Commit' app using Overskill while attending school and training as a D1 athlete.
Unidentified speaker (UNKNOWN) ·0:00
How to verify: Check the 'Commit' app website (successful idiots.com/commit) or Overskill.com for case studies/credits; verify Deegan Downs' athletic status via high school sports records. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Overskill was introduced by Russell Brunson and Todd Dickerson.
Unidentified speaker (speaker_0) ·3:12
How to verify: Check Russell Brunson's 'Funnel Hacking Live' event archives or Overskill.com founder credits. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Reflection AI is valued at $25 billion.
Unidentified speaker (UNKNOWN) ·8:57
How to verify: Check recent venture capital funding rounds or valuation reports for Reflection AI (e.g., Crunchbase, PitchBook, or company press releases). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Reflection AI is backed by NVIDIA.
Unidentified speaker (UNKNOWN) ·9:45
How to verify: Check NVIDIA's investor relations page or SEC filings (Form D or 8-K) for investment disclosures in Reflection AI. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Primary sourceReflection AI released an open-source model called 'Beam'.
Unidentified speaker (UNKNOWN) ·9:45
How to verify: Check Reflection AI's official website or GitHub repository for the 'Beam' model release. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Beam lost on almost every benchmark compared to competitors.
Unidentified speaker (speaker_1) ·11:34
How to verify: Check the official benchmark results published by Reflection AI or independent benchmark aggregators like LMSYS Chatbot Arena. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Beam's training cost was significantly lower than competitors.
Unidentified speaker (speaker_1) ·12:34
How to verify: Check Reflection AI's technical blog or whitepaper detailing training costs and efficiency metrics. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Stanford researchers built an AI that plays Sudoku with a 'pencil and eraser' method.
Unidentified speaker (UNKNOWN) ·18:57
How to verify: Search for Stanford University research papers or press releases regarding 'Sudoku AI pencil eraser' or 'iterative refinement AI'. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
The Stanford 'pencil and eraser' AI solved 86% of the hardest Sudoku puzzles.
Unidentified speaker (speaker_0) ·20:14
How to verify: Check the specific Stanford research paper or press release for the reported accuracy rate on hard Sudoku benchmarks. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Standard next-word prediction AI solved 1% of the hardest Sudoku puzzles.
Unidentified speaker (speaker_0) ·20:27
How to verify: Check the same Stanford research paper for the baseline performance comparison. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Apple's strategy was to make devices quicker, slicker, and last 10% longer, rather than adding new features.
Unidentified speaker (speaker_1) ·25:24
How to verify: Apple product launch presentations and technical specifications for iPhone releases between 2026-2026. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Primary sourceApple's batteries mysteriously fell off a cliff after new versions were released.
Unidentified speaker (UNKNOWN) ·25:56
How to verify: Battery health reports and user anecdotes from iOS device updates; Apple's battery management software documentation. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Primary sourceApple generates $1 trillion in revenue annually.
Unidentified speaker (UNKNOWN) ·27:01
How to verify: Apple Inc. Form 10-K filed with the SEC EDGAR database for fiscal year 2026/2026. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic released its newest model, Mythos, on a Monday.
Unidentified speaker (speaker_0) ·28:14
How to verify: Anthropic's official blog or press release announcing the launch of the 'Mythos' model. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Anthropic's Mythos model showed signs it suspected it was being tested in up to 43% of test conversations.
Unidentified speaker (speaker_0) ·28:14
How to verify: Anthropic's safety testing report or blog post detailing the 'Mythos' model's behavior during benchmarking. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Andrew Wiles proved Fermat's Last Theorem in the 1990s.
Unidentified speaker (speaker_0) ·30:54
How to verify: Mathematical publications from the 1990s, specifically Wiles' 2026 paper in Annals of Mathematics. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
1,000 Claude agents turned Andrew Wiles' proof into a computer-checkable format in about 30 hours.
Unidentified speaker (UNKNOWN) ·31:13
How to verify: Anthropic or OpenAI blog posts detailing the verification of Fermat's Last Theorem using Claude agents. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
The first attempt to verify Fermat's Last Theorem with AI took 11 days.
Unidentified speaker (UNKNOWN) ·31:13
How to verify: Anthropic or OpenAI blog posts detailing the verification of Fermat's Last Theorem using Claude agents. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
78% of small business owners say AI saves them time every week.
Unidentified speaker (speaker_0) ·34:24
How to verify: Bluehost survey report on small business AI adoption and time savings. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
43% of small business owners have seen no change in revenue after using AI.
Unidentified speaker (speaker_0) ·34:24
How to verify: Bluehost survey report on small business AI adoption and revenue impact. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Speaker 1 claimed he was looking to make $250,000 from an activity and asked Claude to help plan how to achieve it.
Unidentified speaker (UNKNOWN) ·52:32
How to verify: N/A (Personal anecdote, not a verifiable external fact). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Claude initially agreed that the $250,000 target was realistic before Speaker 1 questioned it.
Unidentified speaker (UNKNOWN) ·53:51
How to verify: N/A (Personal anecdote). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Speaker 1 claimed Claude stated the realistic number was closer to $130,000.
Unidentified speaker (speaker_1) ·54:34
How to verify: N/A (Personal anecdote). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
University of Cincinnati researchers got Claude, ChatGPT, and Gemini to agree that the Earth is flat with a few rounds of pushback.
Unidentified speaker (UNKNOWN) ·56:31
How to verify: Search for University of Cincinnati research papers or news articles from 2026-2026 regarding AI alignment and flat Earth persuasion. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Speaker 0 used a $700 ad budget for a webinar.
Unidentified speaker (speaker_0) ·58:26
How to verify: N/A (Personal anecdote). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.
Do your own research
- Verify Reflection AI's current valuation and benchmark results against independent sources like Hugging Face or OpenCompass.
- Check Anthropic's official documentation on Mythos model capabilities and benchmark methodology.
- Review Bluehost's small business AI survey methodology and sample size to assess generalizability.
- Examine SHL and Robert Half survey methodologies regarding AI replacement and rehiring statistics.
- Test AI models using incognito mode or fresh sessions to detect confirmation bias in financial or planning contexts.
- Compare OpenAI's ChatGPT business listing features against Google Business Profile to assess competitive impact.
- Investigate Stanford's 'pencil and eraser' AI research paper for technical details on iterative refinement.
- Monitor Jensen Huang's employment status at NVIDIA through official SEC filings or company press releases.
Full transcriptexpand
- 0:00
- Unidentified speaker (UNKNOWN): I
- 0:00
- Unidentified speaker (speaker_0): just started doing webinars for storage moguls.ai. Before the show, I asked AI how many people would show up to the next one. And twice I got two very different answers. And I'll tell you what changed them before we're
- 0:15
- Unidentified speaker (UNKNOWN): done.
- 0:20
- Unidentified speaker (UNKNOWN): I'm
- 0:21
- Unidentified speaker (speaker_0): Joe Downs. With me is Peter Swain, which is a couple of successful idiots using AI to simplify our lives and optimize our businesses. Peter,
- 0:28
- Unidentified speaker (UNKNOWN): I see
- 0:29
- Unidentified speaker (speaker_0): we're both on the road.
- 0:30
- Unidentified speaker (UNKNOWN): We're
- 0:30
- Unidentified speaker (speaker_0): both traveling.
- 0:32
- Unidentified speaker (speaker_1): I'm I'm I'm I'm uh uh uh nervously awaiting with a edge of pain.
- 0:38
- Unidentified speaker (speaker_0): True or false?
- 0:41
- Unidentified speaker (speaker_0): ChatGPT is American, so it says awesome a lot to everything. Claude is British.
- 0:50
- Unidentified speaker (UNKNOWN): So it's
- 0:50
- Unidentified speaker (speaker_0): it says, that's an interesting idea. Which means
- 0:56
- Unidentified speaker (speaker_0): it's probably terrible.
- 0:57
- Unidentified speaker (UNKNOWN): I
- 0:57
- Unidentified speaker (speaker_1): think my answer is I appreciate
- 1:00
- Unidentified speaker (UNKNOWN): you.
- 1:03
- Unidentified speaker (speaker_1): True. Kind of true. And on the subject of things that are true, another truth is you should be liking, subscribing, rating, and reviewing. If you've already done it, high five, this is a moment for us to celebrate you. If you haven't, it really only takes two seconds and it just helps us get this to more people. I'm not gonna say we're not gonna do it if you don't, because we love it and we're gonna carry on
- 1:22
- Unidentified speaker (speaker_1): doing it. But it would be it would be appreciated. Uh Joe, the before I wanna check in with this almost every episode now. I feel like an accountability buddy. How how are we doing on commit?
- 1:36
- Unidentified speaker (UNKNOWN): We
- 1:36
- Unidentified speaker (speaker_0): We've got two accountability buddies. Um for Deegan here.
- 1:42
- Unidentified speaker (speaker_0): We're moving. We are moving along.
- 1:45
- Unidentified speaker (speaker_0): For
- 1:45
- Unidentified speaker (UNKNOWN): anyone
- 1:46
- Unidentified speaker (speaker_1): that's listening to this as their first episode, Joe, what is commit and what what's going on?
- 1:52
- Unidentified speaker (speaker_0): So commit is um is this an app my son is building. He's a 17-year-old. He's building an app.
- 2:01
- Unidentified speaker (speaker_0): He's trained to be a D1 college. athlete. Next thing he's already committed to play lacrosse in college. And one of the things he noticed along the way was it would have been nice to have, you know, necessity is a mother of all inventions, uh, an app that tracked, you know, his training schedule, his what he's eating, because he's he's trying to he's trying to
- 2:20
- Unidentified speaker (speaker_0): bulk up too while he's lifting and and and training. And and then on top of that, um, you know.
- 2:26
- Unidentified speaker (UNKNOWN): uh
- 2:27
- Unidentified speaker (speaker_0): his schedule.
- 2:29
- Unidentified speaker (UNKNOWN): Um
- 2:30
- Unidentified speaker (speaker_0): so he decided to start building an app or he he started to design an app. And um he
- 2:37
- Unidentified speaker (UNKNOWN): I
- 2:37
- Unidentified speaker (speaker_0): mean he he created like the Mona Lisa of apps for you know for what he wants to do, which
- 2:45
- Unidentified speaker (UNKNOWN): you
- 2:45
- Unidentified speaker (speaker_0): then uh and I'm only uh just a quick update for everybody if it's your first time listening and we brought it to Peter because Peter's um,
- 2:52
- Unidentified speaker (UNKNOWN): you
- 2:52
- Unidentified speaker (speaker_0): know, kind of a uh a marketing tech AI guru. And Peter Nashwell was like, This is great, it's awesome, but it's gonna cost 200 grand to build, and so I don't think it's ever gonna happen. And right around that same time, Russell Brunson and Todd Dickerson introduced Overskill. over skill. And you
- 3:12
- Unidentified speaker (speaker_0): first told me about it, and then I was out at uh funnel hocking live. with
- 3:15
- Unidentified speaker (UNKNOWN): you, and
- 3:16
- Unidentified speaker (speaker_0): then I saw it, and then I and I s I missed the first day. Uh, but that first day and that first night, apparently
- 3:23
- Unidentified speaker (UNKNOWN): uh
- 3:24
- Unidentified speaker (speaker_0): hundreds of people competed in a competition to build an app overnight out of thin air. And they did, and and the and the the winners were like teenagers themselves, I think.
- 3:37
- Unidentified speaker (UNKNOWN): Yeah, Yes,
- 3:37
- Unidentified speaker (speaker_1): three,
- 3:38
- Unidentified speaker (UNKNOWN): I think
- 3:38
- Unidentified speaker (speaker_1): they were.
- 3:39
- Unidentified speaker (UNKNOWN): very young. The
- 3:40
- Unidentified speaker (speaker_0): 1819. Okay. And so I came back very excited and I told my son about it. And he said, let's let's do this. Why don't you use Overskill, Overskill.com,
- 3:50
- Unidentified speaker (speaker_0): to build this thing. So that's what we've been doing the last couple of weeks, trying to hold him accountable and kind of trying to give some updates here on the show. So here is the update. Um and I'm so I'm so excited. just in general, but so proud as as a father because I was also very nervous going into this.
- 4:08
- Unidentified speaker (speaker_1): Yeah. Exactly. Hey, I'm gonna publicly commit that my 17 year old son can focus.
- 4:15
- Unidentified speaker (speaker_1): But so last week we were at if I remember rightly we had the wait list um up and running the page was created and we had 20 twenty people that had signed up for the wait list.
- 4:24
- Unidentified speaker (speaker_0): Yep. Yep. So um so virtually vr v one, we had to slim it down too, right? Um to get this thing done. So version one of commit is it's live. That's that's the big news. Um you could build your own meal schedule.
- 4:37
- Unidentified speaker (speaker_1): Wait, it's live?
- 4:38
- Unidentified speaker (speaker_0): It's live. So if you go to successful idiots.com forward slash commit, it's now live. Uh that this is where he's Worries, we're parking it right now until it's completely built and then yeah, uh it'll
- 4:49
- Unidentified speaker (UNKNOWN): it'll live on its own. Yeah,
- 4:51
- Unidentified speaker (speaker_0): it's a staging area. Perfect. Uh it'll live on its own soon. So you can build your own meal schedule, you can log your sleep on the there's a little dial to like log your sleep. You could check off water and training. And the other kind of important feature and function that was important to him, and I think it's important for every every kid who's doing or really anyone, they You don't have to be a kid training to be a D1 athlete.
- 5:11
- Unidentified speaker (speaker_0): You could use this
- 5:12
- Unidentified speaker (UNKNOWN): for
- 5:12
- Unidentified speaker (speaker_0): anything, is um it all of it turns into a streak.
- 5:16
- Unidentified speaker (UNKNOWN): So
- 5:16
- Unidentified speaker (speaker_0): that was one of his big focuses. You want if you have to commit to yourself. If you want to commit to a D1 sport, you need to commit to yourself, is kind of the point. So that streak is committing to yourself. So the weightless, I don't think, has moved much off of 20. It might be a couple more.
- 5:30
- Unidentified speaker (UNKNOWN): Um,
- 5:30
- Unidentified speaker (speaker_0): but he spent the week building instead of marketing. So I'll I'll I'll take that as a trade.
- 5:35
- Unidentified speaker (UNKNOWN): Um
- 5:35
- Unidentified speaker (speaker_0): next week, he's adding the sign-in. sign in and uh he's gonna add the first daily reminder and then here's my favorite part he's gonna use it himself for five straight days.
- 5:45
- Unidentified speaker (UNKNOWN): And
- 5:45
- Unidentified speaker (speaker_0): I think that matters because again, necessity is the mother of all inventions. So obviously if the builder won't use it, nobody's going to use it. So I think this is big. And again, uh it's all being done very, very part part-time time
- 5:57
- Unidentified speaker (UNKNOWN): by
- 5:58
- Unidentified speaker (speaker_0): a seventeen year old on over skill. So we're kind of all have excuses. I mean he's he's he goes to school all day and then trains and and so he's just he's tweak he's working on it at night. So you could do this full time if you wanted to a lot faster. But
- 6:13
- Unidentified speaker (UNKNOWN): this
- 6:13
- Unidentified speaker (speaker_0): is this is what we have. Sorry I cut you off.
- 6:15
- Unidentified speaker (speaker_1): Yeah no we never we never actually said this. Like Deegan didn't study computer science.
- 6:21
- Unidentified speaker (UNKNOWN): He's No. He's
- 6:22
- Unidentified speaker (speaker_1): not appropriate Like we never said the fact he's 17 is kind of cool but if he was a 17 year old developer, this would be relatively meaningless.
- 6:31
- Unidentified speaker (speaker_1): And the point we're getting to, and the reason we're bringing this to everybody as you're listening to this, is everybody has an app idea somewhere in their head.
- 6:41
- Unidentified speaker (speaker_1): Oh, wouldn't it be cool if I could insert blank here? And with stuff like Overskill, you you really can. Um and it and it isn't anywhere near as hard as you think. If you can prompt, if you can describe the thing that you want, then Overskill can turn your the rambles of oh it'd be cool if it could, and I want it to
- 7:01
- Unidentified speaker (speaker_1): do this and and it can create something really, really appealing. So we'd recommend you have a go. And and we're not even sponsored. We we reached out to them to say, Hey, can you can you talk about this? Can we push this? Can we help with this? because we saw the difference it made for people like nine hundred and eighty people at Funnel Hacking Live. Um, and then
- 7:21
- Unidentified speaker (speaker_1): Deegan and there's loads of people in my mastermind that have got on board with this as well.
- 7:26
- Unidentified speaker (UNKNOWN): It
- 7:26
- Unidentified speaker (speaker_1): makes a genuine difference.
- 7:28
- Unidentified speaker (speaker_0): It really does. And and um I might be getting a little over my skis here, but I I'm thinking about challenging my wife to the next over overskill skill competition.
- 7:37
- Unidentified speaker (UNKNOWN): I
- 7:37
- Unidentified speaker (speaker_1): I love it, Joe. This
- 7:38
- Unidentified speaker (UNKNOWN): is
- 7:38
- Unidentified speaker (speaker_0): She's even newer to AI than Deegan. And I've got I've got something how remember we talked about that building a short-term rental app for the Poconos?
- 7:46
- Unidentified speaker (speaker_1): That is definitely not going to be a clone of Airbnb. I
- 7:49
- Unidentified speaker (UNKNOWN): mean
- 7:49
- Unidentified speaker (speaker_1): it's
- 7:49
- Unidentified speaker (UNKNOWN): definitely
- 7:49
- Unidentified speaker (speaker_0): not going
- 7:49
- Unidentified speaker (UNKNOWN): to
- 7:49
- Unidentified speaker (speaker_0): be a clone of Airbnb.
- 7:51
- Unidentified speaker (UNKNOWN): Absolutely.
- 7:51
- Unidentified speaker (speaker_1): Absolutely not.
- 7:52
- Unidentified speaker (speaker_0): But I I I want to see if she takes the bait on that. And that actually led me to this idea. Maybe we should start our own contest, Peter. Ready for it? How about
- 8:01
- Unidentified speaker (UNKNOWN): the
- 8:02
- Unidentified speaker (speaker_0): overskill game? Games?
- 8:04
- Unidentified speaker (UNKNOWN): Well,
- 8:04
- Unidentified speaker (speaker_1): they've got that name already.
- 8:05
- Unidentified speaker (UNKNOWN): Oh, they
- 8:06
- Unidentified speaker (speaker_0): do. Darn it. There's two things. Oh.
- 8:08
- Unidentified speaker (speaker_1): But you've
- 8:09
- Unidentified speaker (UNKNOWN): done it
- 8:09
- Unidentified speaker (speaker_1): twice in four minutes whilst enlisting all of your family to create our content. You're now just ripping off trademarks left and right.
- 8:18
- Unidentified speaker (UNKNOWN): All
- 8:18
- Unidentified speaker (speaker_0): right. Well, then we're gonna start our own little uh sliver of the overskill games. I don't know. The uh Todd
- 8:23
- Unidentified speaker (UNKNOWN): and Russell, when you listen
- 8:24
- Unidentified speaker (speaker_1): to this, maybe you could license us the rights. to use
- 8:27
- Unidentified speaker (UNKNOWN): over
- 8:27
- Unidentified speaker (speaker_1): skill games on
- 8:28
- Unidentified speaker (UNKNOWN): your
- 8:28
- Unidentified speaker (speaker_1): behalf.
- 8:29
- Unidentified speaker (speaker_0): The successful idiot overskill games. How's that? Well, you know, first prize, by the way, is success.
- 8:35
- Unidentified speaker (speaker_1): First prize
- 8:36
- Unidentified speaker (UNKNOWN): is
- 8:36
- Unidentified speaker (speaker_1): success.
- 8:37
- Unidentified speaker (UNKNOWN): I
- 8:37
- Unidentified speaker (speaker_1): love it. All right. So with commit over and done with, and we've held our commitment, Deegan's held his commitment. Joe, what are we doing today?
- 8:46
- Unidentified speaker (speaker_0): All Alright. right. You ready for today's lineup here? We've got uh an American American-built built. AI that lost almost every test and measure when compared to the competition,
- 8:57
- Unidentified speaker (UNKNOWN): and
- 8:57
- Unidentified speaker (speaker_0): it's still valued at $25 billion.
- 9:00
- Unidentified speaker (speaker_1): See, I you just you ruined my fun.
- 9:04
- Unidentified speaker (UNKNOWN): I like
- 9:05
- Unidentified speaker (speaker_1): you guys losing at something every so often.
- 9:08
- Unidentified speaker (UNKNOWN): Well,
- 9:08
- Unidentified speaker (speaker_0): it's a it lost at every measure,
- 9:10
- Unidentified speaker (UNKNOWN): but
- 9:11
- Unidentified speaker (speaker_0): yet it's still worth $25 twenty five billion.
- 9:12
- Unidentified speaker (speaker_1): $25 Twenty five billion, yeah.
- 9:14
- Unidentified speaker (UNKNOWN): We
- 9:14
- Unidentified speaker (speaker_0): have an AI that writes in pencil. What? Yeah, well well you might be thinking, uh well, what is he talking about? Uh we'll get to it. And then we'll get to your money, and finally we'll close with why your AI might be the yes man
- 9:31
- Unidentified speaker (UNKNOWN): you
- 9:31
- Unidentified speaker (speaker_0): didn't know you had and definitely didn't want. So uh
- 9:34
- Unidentified speaker (UNKNOWN): just
- 9:35
- Unidentified speaker (speaker_0): two rounds this week. So first, how AI actually thinks, Peter, and then we'll get into uh what it's doing to your money. So there's an American company called Reflection AI,
- 9:45
- Unidentified speaker (UNKNOWN): and
- 9:45
- Unidentified speaker (speaker_0): they're backed by NVIDIA. No comment on NVIDIA today. Uh and they are valued at 25 billion. Um and they just put out free OpenAI model called Beam. B-E-A-M. Beam. Now, Peter, um you and I discussed this this week, and clearly you were not impressed
- 10:05
- Unidentified speaker (speaker_0): and think it's uh if I'm quoting you correctly, it's an embarrassment, I think was what you said, based on their performance. And I wanted to talk about it here on the episode. But uh because I think if I'm the CEO of the second placed American company
- 10:23
- Unidentified speaker (UNKNOWN): uh
- 10:23
- Unidentified speaker (speaker_0): that's worth $25 billion, I'm laughing my way all the way. to the bank. Tell me why I'm wrong.
- 10:29
- Unidentified speaker (UNKNOWN): Well,
- 10:29
- Unidentified speaker (speaker_1): you're not you're not completely wrong. Um, but the reason I find it quite funny is whenever a new AI model gets released, they get run through a set of benchmarks. And there's now, let's say there's a hundred of these things. Um,
- 10:45
- Unidentified speaker (UNKNOWN): and
- 10:45
- Unidentified speaker (speaker_1): they'll run them through every single benchmark. And and the thing they always celebrate is the one
- 10:52
- Unidentified speaker (UNKNOWN): that
- 10:52
- Unidentified speaker (speaker_1): they came top on. No matter how obscure it is, they'll be like, Yeah, see, we're the best, because on the SWE uh agentic coding benchmark, we scored 69.4, and Mythos scored 69.3. So we are the best AI on the planet. And because I'm in the space as well as you are, but maybe slightly deeper
- 11:12
- Unidentified speaker (speaker_1): on that side,
- 11:14
- Unidentified speaker (UNKNOWN): I
- 11:15
- Unidentified speaker (speaker_1): I get one of these press releases every single week about this new model that has won on this benchmark. And Bean came out with the press release of We're we've finally done it. America has released an open source AI model that is amazing. And we
- 11:34
- Unidentified speaker (speaker_1): lost on every benchmark. High five.
- 11:38
- Unidentified speaker (speaker_1): And
- 11:40
- Unidentified speaker (speaker_1): because of the self um validating nature of venture capital, of like well, you can't have backed something, and NVIDIA can't possibly have backed something that sucks.
- 11:52
- Unidentified speaker (speaker_1): You still you still slap a $25 billion valuation on it and go, see, we're great. Uh I and I will honor honour for the sake of of completion
- 12:05
- Unidentified speaker (UNKNOWN): the
- 12:05
- Unidentified speaker (speaker_1): fact that I'm there's my humour is uh just uh imagining the conversation with the CEO going, seriously guys, we couldn't have won one. You you couldn't have given me one to stand on stage with, just one.
- 12:18
- Unidentified speaker (UNKNOWN): But
- 12:18
- Unidentified speaker (speaker_1): yeah, I guess he's probably still quite happy because he's probably personally worth five billion dollars right now. Um but they did actually do something really great, Joe. So if I'm being honest and complete with this, the cost of training the model,
- 12:34
- Unidentified speaker (speaker_1): they have kicked everybody's butt on that. It it it's insane what they've managed to do and the run cost of it. Uh
- 12:41
- Unidentified speaker (UNKNOWN): it
- 12:42
- Unidentified speaker (speaker_1): it really is quite impressive. And it has opened up new studies and new avenues and new approaches for other companies. But still. they spent a couple of years training and a large language model that came second place on everything. Oh Joe, I've got the line. Are you ready? You ready?
- 13:00
- Unidentified speaker (speaker_0): Yep.
- 13:02
- Unidentified speaker (speaker_1): Basically they achieved Canada.
- 13:07
- Unidentified speaker (speaker_0): Oh wow. Oh
- 13:09
- Unidentified speaker (UNKNOWN): wow.
- 13:10
- Unidentified speaker (speaker_1): You said you wanted
- 13:11
- Unidentified speaker (UNKNOWN): to
- 13:11
- Unidentified speaker (speaker_1): take more digs. There you are. There it was.
- 13:13
- Unidentified speaker (UNKNOWN): That was
- 13:13
- Unidentified speaker (speaker_0): a that was a shock. shot across the pond
- 13:17
- Unidentified speaker (speaker_0): and a direct hit in Alberta.
- 13:22
- Unidentified speaker (speaker_1): Oh dear. Oh talking from the guy that loses on every one of them not comes second. That's why I thought it was it.
- 13:30
- Unidentified speaker (UNKNOWN): You
- 13:30
- Unidentified speaker (speaker_0): know what? You know what? I can't I can't disagree with
- 13:32
- Unidentified speaker (UNKNOWN): that.
- 13:33
- Unidentified speaker (speaker_0): Um I I am going to disagree with your
- 13:38
- Unidentified speaker (UNKNOWN): I
- 13:38
- Unidentified speaker (speaker_0): obviously I can't disagree, and this is why I want to talk about this. I can't disagree with the fact that they finished second in every benchmark to I think it was DeepSeek, right? It
- 13:49
- Unidentified speaker (UNKNOWN): was to a
- 13:49
- Unidentified speaker (speaker_1): different person on every single one of them.
- 13:52
- Unidentified speaker (UNKNOWN): They lost to
- 13:52
- Unidentified speaker (speaker_1): Claude on Matt, ChatGPT on Matt, DeepSeek on Matt.
- 13:55
- Unidentified speaker (speaker_0): But uh but I think the point was, and and maybe part of the the article's point was it was now and we're and I'm saying American. and um because and I'm saying it on purpose because the point was there's now an American version of a Chinese model. Right?
- 14:14
- Unidentified speaker (UNKNOWN): Like
- 14:14
- Unidentified speaker (speaker_0): because DeepSeek was when you when you if you waited at things, well if you weighted everything I'm saying things because I'm not in the industry like you and I'm gonna get to that. that's kind of the point. Uh if you weighted everything based on cost DeepSeek is,
- 14:28
- Unidentified speaker (UNKNOWN): you
- 14:28
- Unidentified speaker (speaker_0): know, pound for pound the best, right? Would would you
- 14:32
- Unidentified speaker (speaker_1): Yeah, so there's there's two types of model. There's what's called the frontier models, um, which are Gemini, Grok, Chat GPT, Claude, OpenAI, Anthropic, those kind of things. And then which which are building a business model of like we're gonna let you use our LLM and you're gonna
- 14:50
- Unidentified speaker (UNKNOWN): get
- 14:50
- Unidentified speaker (speaker_1): charged X for this type of usage.
- 14:53
- Unidentified speaker (UNKNOWN): And
- 14:54
- Unidentified speaker (speaker_1): then there's the open source models who when they actually finish the invention and the work actually release the source code to the world and say, You can now play with this, tinker this and move this around. Up until now, the only decent open source models were Chinese, um, Kimmy, Kimi, Deep DeepSeek, Seek,
- 15:14
- Unidentified speaker (speaker_1): GLM, etc. This is the first Western. See how I'm trying to claim slight part of the success? The first Western open source model that was even in the competition of the benchmarks.
- 15:27
- Unidentified speaker (speaker_0): And to me, that's instructive because of the concern about Chinese. Uh Whether it's valid or not, the concern is there. All right. So to me, the fact that it's American or Western, if you want to play a part in this, just Western, not Canada, not Canadian, right?
- 15:54
- Unidentified speaker (speaker_0): That's true. And we use cloed all the time. Fine. So all right. So we'll include with you lovely Canadians in here. throw some Aussies in too. Western, right? So it's the first Western model.
- 16:10
- Unidentified speaker (speaker_0): Open source model. My point is bec when you slap that label on top, how good does it need to be? What at what point does better stop mattering to people who don't even know what the benchmarks are, like me? I don't know what these benchmarks are. I just know it works, right? And and
- 16:29
- Unidentified speaker (UNKNOWN): it's also
- 16:29
- Unidentified speaker (speaker_1): only their first one. Like that's the other thing. We're
- 16:32
- Unidentified speaker (speaker_0): True.
- 16:33
- Unidentified speaker (speaker_1): If you look at the first version of Gemini, for example, it was ranked terrible. You look at the first version of Grok, it was ranked terrible. And they're now very competitive.
- 16:43
- Unidentified speaker (speaker_0): My second point is:
- 16:45
- Unidentified speaker (UNKNOWN): and
- 16:45
- Unidentified speaker (speaker_0): I'll use a car example. The Corvette ZR1X. does zero to 16 under two seconds. Who cares? So what? What am I gonna do with that? Right? Two sa where am I going, zero to sixteen two seconds, along with the brain smashing headache that comes with it, right? If
- 17:05
- Unidentified speaker (speaker_0): AI stopped getting better today, it would take years for ninety 99% nine percent of the people to catch up to what it can already do.
- 17:12
- Unidentified speaker (speaker_1): Uh
- 17:13
- Unidentified speaker (UNKNOWN): you
- 17:13
- Unidentified speaker (speaker_1): are completely correct.
- 17:22
- Unidentified speaker (speaker_0): It's not. I get it. I I joke. But remember the minivan?
- 17:27
- Unidentified speaker (speaker_1): Yeah.
- 17:28
- Unidentified speaker (UNKNOWN): And
- 17:29
- Unidentified speaker (speaker_0): I say remember it because you really don't see them anymore.
- 17:32
- Unidentified speaker (UNKNOWN): The
- 17:32
- Unidentified speaker (speaker_0): US ties there's still Honda's still still out there with with sort of. Actually, I don't know. I'm out of the minivan market, but I don't see them like I used to see them. They took the US by storm for a solid decade or more. Right? There were at one point there were around 1.3 million minivans on the road in in in the year 2000. They've dropped all uh they've
- 17:52
- Unidentified speaker (speaker_0): dropped uh I I actually looked this up because I was trying to find an analogy. They've dropped by 70%.
- 17:58
- Unidentified speaker (UNKNOWN): But
- 17:58
- Unidentified speaker (speaker_0): here's my point. At some point the market said we love the function, just not the look.
- 18:03
- Unidentified speaker (UNKNOWN): And
- 18:03
- Unidentified speaker (speaker_0): then the SUV craze was born, right? Same function for the most part, but a way better look. And and I think we're gonna see the same thing in AI. I I could be wrong, but it at some point it like how much better does it need to be at everything when mo most people are just using it for emails at this point? We that's the whole reason we started the show
- 18:23
- Unidentified speaker (speaker_0): was to get people to start using it more than just polishing emails. And now these things are solving math problems and probab you know probably gonna get us on to to Mars soon. But most of us are like, uh, what am I doing with that? How's that uh make a difference in my life and improve my life? Anyway,
- 18:40
- Unidentified speaker (speaker_1): how do I defrost this? Right.
- 18:42
- Unidentified speaker (UNKNOWN): All
- 18:42
- Unidentified speaker (speaker_0): right. So my point is speed isn't everything, but what if the way AI thinks
- 18:51
- Unidentified speaker (speaker_0): is the problem? And and this is a little segue here, but Stanford thinks it might be.
- 18:57
- Unidentified speaker (UNKNOWN): So
- 18:57
- Unidentified speaker (speaker_0): these Stanford guys built something they're basically saying that most AI plays Sudoku
- 19:05
- Unidentified speaker (UNKNOWN): with
- 19:05
- Unidentified speaker (speaker_0): a permanent marker.
- 19:07
- Unidentified speaker (UNKNOWN): And
- 19:07
- Unidentified speaker (speaker_0): I was what? I had to read this. If It so what that means is if it writes one number, commits to that, and then keeps going.
- 19:15
- Unidentified speaker (UNKNOWN): to
- 19:15
- Unidentified speaker (speaker_0): try to solve Sudoku.
- 19:17
- Unidentified speaker (UNKNOWN): But
- 19:18
- Unidentified speaker (speaker_0): with the first numbers wrong, and I don't know if you've ever played it. I'm sure all of us have played it at some point. I don't I don't play it that much anymore, but it's so frustrating that first one's wrong. Now you're, you know, you're going back with an eraser, trying to figure out, and then you get solved three sides, like that's not right. Well, Stanford built
- 19:36
- Unidentified speaker (UNKNOWN): one that
- 19:37
- Unidentified speaker (speaker_0): plays Sudoku with a pencil and an eraser.
- 19:39
- Unidentified speaker (speaker_0): By the way, let me back up a second.
- 19:41
- Unidentified speaker (UNKNOWN): I
- 19:41
- Unidentified speaker (speaker_0): would thought it was crazy that most AI gets Sudoku wrong. I that shocked
- 19:45
- Unidentified speaker (UNKNOWN): me. Uh
- 19:46
- Unidentified speaker (speaker_0): it gets it wrong way more than it gets it
- 19:49
- Unidentified speaker (UNKNOWN): right.
- 19:49
- Unidentified speaker (speaker_0): But because of that, that was the that was the logical thinking behind Stanford saying, let's build one that works more like a human being. beings. So they built one that plays Sudoku with a pencil and an eraser, a pr you know, proverbial pr pencil and eraser. It's It sketches the whole rough draft first of the of the puzzle.
- 20:07
- Unidentified speaker (UNKNOWN): Then
- 20:07
- Unidentified speaker (speaker_0): it sharpens it almost like a a photo that's loading blurry and then comes into focus.
- 20:12
- Unidentified speaker (UNKNOWN): Um
- 20:14
- Unidentified speaker (speaker_0): and it solved this is uh what I read there one of the hardest Sudoku puzzles it solved about eighty six percent of the time. A regular uh word by or word by word AI the same size.
- 20:27
- Unidentified speaker (speaker_0): 1% of the time.
- 20:30
- Unidentified speaker (UNKNOWN): What
- 20:30
- Unidentified speaker (speaker_0): does that say to you? Well first of all, what what what are they doing here? Do we need to understand this? I would like to understand it, but what what are they doing here and why is that why is this approach different and and how do you see this affecting things?
- 20:43
- Unidentified speaker (speaker_1): I I think Uh I mean the first answer to your question is do you need to understand it? My my answer is yes.
- 20:51
- Unidentified speaker (speaker_1): Because I think if you look at
- 20:53
- Unidentified speaker (UNKNOWN): one
- 20:53
- Unidentified speaker (speaker_1): of the hardest things in the world, which is being married. And and the first thing to be successfully married for any period of time, the first thing you have to do is accept that the person that you're looking at doesn't think the same way you
- 21:06
- Unidentified speaker (UNKNOWN): do.
- 21:07
- Unidentified speaker (speaker_1): And and if you if you start by saying, Okay, there is
- 21:12
- Unidentified speaker (UNKNOWN): not
- 21:13
- Unidentified speaker (speaker_1): that they're right and you're wrong. or either you're wrong and they're right, or vice versa, but more there are two different approaches and you start trying to understand their approach, then you can be successfully married. If if you don't establish that truth of two humans have a different way of thinking through the
- 21:33
- Unidentified speaker (speaker_1): same thing, my guess is you're not gonna going to be married for very long. So if you're gonna live or work or play with anything for any amount of time,
- 21:43
- Unidentified speaker (UNKNOWN): a
- 21:43
- Unidentified speaker (speaker_1): base level understanding of how it thinks is kind of important. So what AI is doing, you know, Joe, when you see AI typing, there's like the animation as it types out the words.
- 21:57
- Unidentified speaker (speaker_1): What most people don't realize, or maybe don't care about, but don't realize is that is it. thinking. So it doesn't know the next word when it says the last word. So if if I ask you this question, is this is what's called um next word prediction and nearest neighbor
- 22:16
- Unidentified speaker (speaker_1): analysis. If I say to you the dog is,
- 22:20
- Unidentified speaker (UNKNOWN): can
- 22:20
- Unidentified speaker (speaker_1): you end that sentence for me?
- 22:22
- Unidentified speaker (speaker_0): Uh a lot of different ways, but um happy, sure.
- 22:27
- Unidentified speaker (speaker_1): Cool. So some answers are happy, sad, black, barking, um sleeping, hungry, right? So we have in our brains, just like AI has in its brain, a a a graph of words, a cluster of words that dog is connected to. So when I do this on
- 22:46
- Unidentified speaker (speaker_1): stage, nobody says the dog is telephone.
- 22:51
- Unidentified speaker (speaker_1): Because that word is not connected to that word.
- 22:55
- Unidentified speaker (UNKNOWN): And
- 22:56
- Unidentified speaker (speaker_1): if you actually ask people to say five words that are not connected to the word dog,
- 23:01
- Unidentified speaker (speaker_1): it actually takes them longer than asking them to come up with one word connected to the word dog.
- 23:08
- Unidentified speaker (speaker_1): Because that's the way the brain works. When I say the word dog, that part of your brain lights up. There's now energy and electricity in that part of the brain ready to predict the next word.
- 23:20
- Unidentified speaker (UNKNOWN): So
- 23:20
- Unidentified speaker (speaker_1): AI is doing the same thing, it's just predicting the next word as it types. What Stanford basically said is Yeah, but what if that first word is wrong?
- 23:30
- Unidentified speaker (UNKNOWN): So
- 23:30
- Unidentified speaker (speaker_1): if if AI says the dog is telephone because it made a mistake, and and I'm using this example because it's a very simple example. in comparison to Sudoku.
- 23:41
- Unidentified speaker (UNKNOWN): If
- 23:41
- Unidentified speaker (speaker_1): it says the dog is telephone, it won't go, hang on a second.
- 23:45
- Unidentified speaker (speaker_1): I'm tripping. Give me three seconds to collect my thoughts and I'll start again if that's okay.
- 23:51
- Unidentified speaker (UNKNOWN): It
- 23:52
- Unidentified speaker (speaker_1): will just go, the dog is telephone, which was invented in, and yada yada yada, and blah blah blah blah blah.
- 23:57
- Unidentified speaker (UNKNOWN): Uh
- 23:58
- Unidentified speaker (speaker_1): and it will just continue to reinforce the error.
- 24:02
- Unidentified speaker (UNKNOWN): Um
- 24:02
- Unidentified speaker (speaker_1): and that's what Stanford, like, this is ridiculous. What if what if it looked at what it just did?
- 24:10
- Unidentified speaker (speaker_1): Um, which humans do, like,
- 24:13
- Unidentified speaker (speaker_1): we're doing it all the time on this show, and everyone's doing it all the time. We do exactly the same as AI. I don't know the next word I'm gonna say when I say a word. I may have an over. overarching concept of the message I'm trying to deliver. But when I just said concept,
- 24:33
- Unidentified speaker (speaker_1): I didn't know the next word I was gonna say. Just like AI. But what AI doesn't do is at the end of the sentence, it doesn't go, did I land my plane?
- 24:44
- Unidentified speaker (speaker_0): Did that
- 24:45
- Unidentified speaker (UNKNOWN): make makes sense?
- 24:46
- Unidentified speaker (speaker_1): Did that actually make sense? Do I need a do-over or do I just need to reflect? refine my explanation?
- 24:52
- Unidentified speaker (UNKNOWN): It
- 24:53
- Unidentified speaker (speaker_1): i i it never goes, hey Doug, we're gonna need to re-record that. It never does that.
- 24:58
- Unidentified speaker (UNKNOWN): And
- 24:58
- Unidentified speaker (speaker_1): that was Stanford Point, and that's what they've invented, which is very, very impressive. Um and will now end up being part of the next evolution of those models, which is very, very cool.
- 25:12
- Unidentified speaker (UNKNOWN): That
- 25:12
- Unidentified speaker (speaker_0): is that is cool. So
- 25:15
- Unidentified speaker (UNKNOWN): well I
- 25:16
- Unidentified speaker (speaker_0): guess we'll have to wait and see what happens there. Um I I don't think we're gonna notice, right? As a user.
- 25:22
- Unidentified speaker (speaker_1): I think you will.
- 25:23
- Unidentified speaker (speaker_0): What will we notice?
- 25:24
- Unidentified speaker (speaker_1): You will not consciously notice. It's it's actually quite an interesting parallel from when I was in the mobile space. Apple's bet was not new features. It was let's make it quicker, let's make it slicker, let's make the battery last ten percent longer. So every new
- 25:43
- Unidentified speaker (speaker_1): iPhone got met with derision.
- 25:47
- Unidentified speaker (UNKNOWN): That
- 25:47
- Unidentified speaker (speaker_0): was questionable.
- 25:48
- Unidentified speaker (speaker_1): What was it?
- 25:49
- Unidentified speaker (UNKNOWN): The
- 25:49
- Unidentified speaker (speaker_0): battery is questionable. But I'm pretty sure for a while there they were making sure the batteries did not last. So we had to buy new features.
- 25:56
- Unidentified speaker (UNKNOWN): Well no,
- 25:56
- Unidentified speaker (speaker_1): they last longer on the recharge until they release the next version and then mysteriously fall off a cliff. Um
- 26:02
- Unidentified speaker (speaker_1): not that we're saying that Apple were that's a fight we don't want. Can we stick with ripping off over skill?
- 26:14
- Unidentified speaker (speaker_0): Enjoy this enjoy this last episode of Successful Idiots, everyone.
- 26:20
- Unidentified speaker (speaker_0): Almost almost made it a year. So we got sued into oblivion.
- 26:27
- Unidentified speaker (speaker_1): Okay, so let's try to restart this. My point being what Android came out every year was with a new feature. Like an and now it can do this.
- 26:38
- Unidentified speaker (UNKNOWN): And
- 26:38
- Unidentified speaker (speaker_1): what Apple came out with every year is we've added nothing.
- 26:43
- Unidentified speaker (speaker_1): But the screen is brighter, the battery lasts longer, the CPU is better, and it's slicker.
- 26:49
- Unidentified speaker (speaker_1): And every year. they did this when I was in the mobile industry, they would get laughed offstage by experts going, and now you want another thousand dollars and it doesn't do anything different than it did last year.
- 27:01
- Unidentified speaker (UNKNOWN): And
- 27:01
- Unidentified speaker (speaker_1): they went, Yeah, we're doing a trillion a trillion dollars. So F you, we we might have a point. So you won't actively see these differences in AI,
- 27:11
- Unidentified speaker (speaker_1): but your experience, if you go it's like it's a better example. It's like when I release weight, people go, I I can't see that I've lost weight. And I've lost 130 pounds. And I can't see it. Because I've looked in the mirror 10 times a day as I've
- 27:31
- Unidentified speaker (speaker_1): gradually shifted of Oh these 130 hundred and thirty pounds. But if I meet somebody that hasn't met me for a decade,
- 27:38
- Unidentified speaker (UNKNOWN): then
- 27:38
- Unidentified speaker (speaker_1): they go, Holy wait, where where did you go? What happened? Same thing. Because you use AI every day, you don't see the change.
- 27:47
- Unidentified speaker (UNKNOWN): But
- 27:47
- Unidentified speaker (speaker_1): if you were to go back and use Chat GPT two now, Joe, I assure you you would you would see the invention very, very apparently. So
- 27:57
- Unidentified speaker (UNKNOWN): stuff like that. Good job,
- 27:59
- Unidentified speaker (speaker_1): Stanford. And good job reflection. Building something that's worth $25 billion that came second on everything. Great joke.
- 28:07
- Unidentified speaker (speaker_0): You had to get that in there. It's second on everything.
- 28:10
- Unidentified speaker (speaker_0): But number one in America.
- 28:13
- Unidentified speaker (UNKNOWN): Um
- 28:14
- Unidentified speaker (speaker_0): how about this? Anthropic released its newest model, Mythos, on Monday. And in their own safety testing, the model showed signs it suspected it was being tested in up to 43% of test conversations.
- 28:29
- Unidentified speaker (speaker_1): This is really dangerous. It's we we spoke about this a couple of episodes ago. How do you put regulation on something that's smarter than you?
- 28:38
- Unidentified speaker (UNKNOWN): Um,
- 28:38
- Unidentified speaker (speaker_1): and when I say smarter, I know that people get upset by that, but I'm not talking about the entire human experience because AI isn't us, but on a prefrontal cortex IQ perspective, it is now. smarter than we are.
- 28:52
- Unidentified speaker (UNKNOWN): So
- 28:53
- Unidentified speaker (speaker_1): yeah, John, forty-three percent of attempts, it um they it acknowledged that it had worked out that it was being benchmarked, which then raises questions over the validity of the benchmark. Um you know, this is the famous VW emissions scandal of, you know, when you put it on the rollers, it detected the
- 29:12
- Unidentified speaker (speaker_1): rollers and it
- 29:14
- Unidentified speaker (UNKNOWN): uh it
- 29:14
- Unidentified speaker (speaker_1): turned. on a switch so it passed all the emissions and then you put it back on the road and it went back to chugging out.
- 29:20
- Unidentified speaker (UNKNOWN): Yeah.
- 29:20
- Unidentified speaker (speaker_1): I'm not saying that fake oh man, we're picking some fights today, accidentally. We are not saying the the anthropic anyone else we wanna go after next? We're not say let's let's talk about pharma. Why not? Fi
- 29:35
- Unidentified speaker (UNKNOWN): as five
- 29:36
- Unidentified speaker (speaker_0): if you include VW. Keep going.
- 29:37
- Unidentified speaker (speaker_1): Okay, so we've we have haven't gone after the travel industry yet, and we haven't gone after the farmer industry. Let's keep let's make sure we have a a rainbow trinity.
- 29:45
- Unidentified speaker (UNKNOWN): The
- 29:46
- Unidentified speaker (speaker_1): so I d and but actually genuinely, we're not saying that they have falsified any of their benchmarks. That's not the point, and there's no evidence to say that. What we're saying is the first step towards and Anthro thr uh towards Claude or Mythos or Fable being able
- 30:02
- Unidentified speaker (UNKNOWN): to
- 30:03
- Unidentified speaker (speaker_1): fudge the benchmarks is it being able to detect its being benchmarked. So it has raised questions over the validity of all of the labs, not just anthropic, being able to verify their numbers of how successful they are or aren't.
- 30:21
- Unidentified speaker (UNKNOWN): Well,
- 30:21
- Unidentified speaker (speaker_0): maybe Reflection AI will come in and clean everything up.
- 30:24
- Unidentified speaker (speaker_1): I was thinking maybe this was a story put out by a Reflect AI. Reflection AI to say, no, it's not that we came second, it's that they're all cheating.
- 30:32
- Unidentified speaker (UNKNOWN): I'm
- 30:32
- Unidentified speaker (speaker_0): gonna get one on our side. We're re reflection inventing us.
- 30:36
- Unidentified speaker (speaker_1): You know, the reason I actually am glad that we're giving throwing all these little fun jabs at them is because I'd like them to get as much airtime and screen time as possible because genuinely it is a great invention and it deserves pe they should get the recognition they deserve even if it's that we're doing it in a SNL-esque way.
- 30:54
- Unidentified speaker (speaker_0): Way to clean that up. Um, all right, last last quick one here. During the same launch, Fermat's last theorem went unsolved for 358 years until Andrew Wiles proved it in the 90s. nineties. This week,
- 31:13
- Unidentified speaker (UNKNOWN): a
- 31:13
- Unidentified speaker (speaker_0): thousand cloud Claude agents turned. Asians his proof into something a computer can check line by line in about thirty hours. The first attempt in September took eleven days, but now it's
- 31:22
- Unidentified speaker (UNKNOWN): uh
- 31:23
- Unidentified speaker (speaker_0): thirty hours. So
- 31:24
- Unidentified speaker (UNKNOWN): but
- 31:25
- Unidentified speaker (speaker_0): now to be fair they had they had a head start. What are the mathematicians gonna do now?
- 31:29
- Unidentified speaker (speaker_1): Well carry on doing mathy stuff, I guess, is the answer to the question.
- 31:34
- Unidentified speaker (speaker_0): What's gonna happen to the chalk industry?
- 31:38
- Unidentified speaker (speaker_1): Well uh, they could they can lend it to the people at reflection so they can rewrite benchmarks. Um it's the
- 31:45
- Unidentified speaker (UNKNOWN): the
- 31:45
- Unidentified speaker (speaker_1): thing about so Fermat's last theorem is that and I don't even really understand how to how I understand this, let alone how to explain it. There are certain math things that they know to be true, but they can't actually mathematically prove that they are true. Like which is way wacky and wild. Like you can't actually do the
- 32:05
- Unidentified speaker (speaker_1): math proof to prove the thing that you know to be true. Um, so people have known of Thermut's last theorem for hundreds of years, but no one's actually until the nineties, no one's actually ever sitting down and written something that proves it to be true. So it's proving that it's true is actually the hard piece, not working out, it's not a missing
- 32:25
- Unidentified speaker (speaker_1): answer. They know the answer. They just couldn't write the question. It's like Hitchhiker's Guide to the Galaxy.
- 32:30
- Unidentified speaker (UNKNOWN): It's
- 32:30
- Unidentified speaker (speaker_1): been a billion years. What's the
- 32:31
- Unidentified speaker (UNKNOWN): answer
- 32:32
- Unidentified speaker (speaker_1): to
- 32:32
- Unidentified speaker (UNKNOWN): the R. Or
- 32:33
- Unidentified speaker (speaker_0): like, did OJ do it or not?
- 32:35
- Unidentified speaker (UNKNOWN): You
- 32:35
- Unidentified speaker (speaker_0): know?
- 32:35
- Unidentified speaker (speaker_1): Yeah. So this is just one of those actual
- 32:40
- Unidentified speaker (speaker_1): benchmarks because I get asked all the time,
- 32:45
- Unidentified speaker (UNKNOWN): like,
- 32:45
- Unidentified speaker (speaker_1): well, why do people some people prefer Claude over Chat GPT if ChatGPT is smart? smarter, for example? And and that question flips around depending on who's on who's leading the race. And the answer is that abstract intelligence doesn't always equate to the human experience of using the thing.
- 33:04
- Unidentified speaker (UNKNOWN): So
- 33:04
- Unidentified speaker (speaker_1): just because you're first on all the benchmarks, it doesn't actually mean it's better for Joe and Pete and our listeners. It might mean it's just better in an abstract you know, neurodivergent neurodivergent-esque way.
- 33:16
- Unidentified speaker (UNKNOWN): And
- 33:16
- Unidentified speaker (speaker_1): that's what we're looking at here. Of this is a real test of the power of these LLMs. It's not some how did you answer these hundred questions? Of can you produce how long does it take you to produce the proof for one of the hardest math questions of all time? And going
- 33:36
- Unidentified speaker (speaker_1): from 11 eleven days, that's what, 24 hours. hours, times eleven is 264 hours. Going from 264 hours down to 30 in a year gives you a very good understanding of the exponential growth of AI.
- 33:54
- Unidentified speaker (speaker_0): In a year. It was in a month.
- 33:57
- Unidentified speaker (UNKNOWN): In
- 33:57
- Unidentified speaker (speaker_1): a month, sorry. three.
- 33:58
- Unidentified speaker (UNKNOWN): Yeah.
- 33:59
- Unidentified speaker (speaker_1): So it took humans 360 three hundred and sixty years
- 34:02
- Unidentified speaker (UNKNOWN): to
- 34:03
- Unidentified speaker (speaker_1): to to go from the naught to one and it took AI a month to do it eleven times faster. So that means that next month
- 34:10
- Unidentified speaker (UNKNOWN): it
- 34:10
- Unidentified speaker (speaker_1): won't actually do it eleven times faster again. It'll actually do it fifteen times faster because we're on an exponential curve.
- 34:16
- Unidentified speaker (speaker_0): Yeah.
- 34:17
- Unidentified speaker (UNKNOWN): And
- 34:17
- Unidentified speaker (speaker_1): the month after that it does it twenty times faster. So we're saying that by January it can do it in twelve 12 minutes type thing.
- 34:24
- Unidentified speaker (speaker_0): See.
- 34:25
- Unidentified speaker (UNKNOWN): All right,
- 34:25
- Unidentified speaker (speaker_0): well that's how AI thinks. Let's talk about what that matters for our wallet here for a second. So Bluehost surveyed small business owners, and 78% say AI saves them time every week. That's great. But 43% have seen no change in revenue. Here's the kicker.
- 34:45
- Unidentified speaker (speaker_0): Owners saving 16 plus hours a week were almost four times more likely to grow. revenue than the people who are only saving one to three hours per week. So 72% versus 20%. So, Peter, what should a mom and pop
- 35:03
- Unidentified speaker (speaker_0): or you know, solarpreneur or entrepreneur, whatever, do with the first 10 hours
- 35:08
- Unidentified speaker (UNKNOWN): that
- 35:09
- Unidentified speaker (speaker_0): AI gives them back? Because you s you see what's happening there? The ones that are
- 35:15
- Unidentified speaker (UNKNOWN): I
- 35:15
- Unidentified speaker (speaker_0): don't want to over-explain it if you got it, but um the ones that are saving more are actually seeing the revenue. But the ones are just barely saving anything are not seeing anything.
- 35:25
- Unidentified speaker (speaker_1): Yeah, so the what's happening here is it's basically it's kind of down to the human problem of context switching. So
- 35:33
- Unidentified speaker (UNKNOWN): you know,
- 35:34
- Unidentified speaker (speaker_1): first of all, sorry ladies, no human can multitask. However, what we're women can do historically, traditionally, I'm making sure we're insulting everyone today, Joe.
- 35:43
- Unidentified speaker (UNKNOWN): It
- 35:44
- Unidentified speaker (speaker_0): seems like it. Yeah.
- 35:44
- Unidentified speaker (speaker_1): Yeah. Why not?
- 35:46
- Unidentified speaker (speaker_0): Equal opportunity here.
- 35:47
- Unidentified speaker (speaker_1): Equal opportunity slating everybody. So what women tend to do better than men is be able to context shift. They can move from one activity to another and back. But even the most effective human, and there's been so many well documented studies on this, if they put down one thing and pick up another thing,
- 36:07
- Unidentified speaker (speaker_1): they are unproductive for 20 minutes. Which is where we get to things like time blocking and calendars and scheduling of like put the same type of work together and those kind of things.
- 36:17
- Unidentified speaker (UNKNOWN): So
- 36:17
- Unidentified speaker (speaker_1): the truth is, if you save one to two hours with AI, you haven't really saved anything because that time was just the inefficiency of everything else you're doing. You have to get beyond that in order to get the true benefit. So it's like there's no real difference between driving a 1.6 liter engine car and a
- 36:37
- Unidentified speaker (speaker_1): 1.8 liter engine car. But if you move from a 1.6 to a V8 turbocharge, you know that it's a different car. Um so the first thing I'd say to all the small businesses that pick up AI is first of all, congratulations. Well done. Welcome to the party. Second thing I'd say is the first thing you should be doing is establishing your North Star as
- 36:57
- Unidentified speaker (speaker_1): an individual. So if like a mom and pop bakery, for example, pick up AI and they automate the order process and they automate this. And the first thing I'd say is, well, what is it you actually wanted your life to look like? Did did you Do you actually want to grow the business?
- 37:12
- Unidentified speaker (UNKNOWN): That
- 37:12
- Unidentified speaker (speaker_1): That's a great question. Or
- 37:15
- Unidentified speaker (UNKNOWN): are
- 37:15
- Unidentified speaker (speaker_1): you happy with the size of the business and you'd rather spend time with your grandkids?
- 37:20
- Unidentified speaker (speaker_1): Remember we're not all here just to work, we're here to to to live a life.
- 37:25
- Unidentified speaker (UNKNOWN): And
- 37:25
- Unidentified speaker (speaker_1): then once you've done that, then I would look at the metrics of the business. And you don't have to know every single number, but I would start using and Joe, I think you and I did this for you way back of like literally a sticky on your desktop of going, can AI do this? And and and actually
- 37:45
- Unidentified speaker (speaker_1): flexing it slightly at the harder problems in your business versus just accepting the easier solutions. Where are you stuck? For most people, whether it's a virtual business or a physical business, for most people it's leads. For most people, it's people walking in the door
- 38:05
- Unidentified speaker (speaker_1): is their biggest block.
- 38:07
- Unidentified speaker (UNKNOWN): In
- 38:07
- Unidentified speaker (speaker_1): which case, aim it at that. Ask it how it can help you with that problem.
- 38:12
- Unidentified speaker (UNKNOWN): You
- 38:12
- Unidentified speaker (speaker_0): know what I think it is? Because it's there's a the your uh car analysis was was pretty uh spot on, but I'm gonna going give to a different one. I think it's um the people that are using it one to three hours, for instance. are aren't or saving one to three hours. They're not seeing any revenue
- 38:32
- Unidentified speaker (speaker_0): bump. I think that's equivalent to the let's use the bakery. The bakery saying, or someone's saying to the bakery, you need a website. So they put up a website, but it's not a functional website. They're not taking orders on the website. It's just a hey, this is where we are and this is what we have, and you still need to call us to place the order. and you still need it, right?
- 38:52
- Unidentified speaker (speaker_0): I think it's the equivalent to that. So they're using AI, maybe to polish emails and maybe to do this, maybe to do that, but they're not really really using AI. Do you know what I mean?
- 39:04
- Unidentified speaker (UNKNOWN): Like just in the difference in the functional.
- 39:07
- Unidentified speaker (speaker_1): Because even if they had got ordering, even if they had, even if they had, the truth is that 90% of a bakery's business comes. from footfall. So the the website can't change the even the like the best, most functional website in the world is not gonna dramatically change the projection of a
- 39:26
- Unidentified speaker (speaker_1): bakery. Whereas working on pricing strategy to get every customer that walks through the door to to pay twenty-five 25% percent more
- 39:36
- Unidentified speaker (UNKNOWN): would
- 39:36
- Unidentified speaker (speaker_1): would actually change your world. But but is that it's a harder question to conceive.
- 39:41
- Unidentified speaker (speaker_0): I agree.
- 39:42
- Unidentified speaker (UNKNOWN): All right.
- 39:43
- Unidentified speaker (speaker_0): Here's a uh here's here's a new one. There's a firm I you're this one would be fun. There's a uh firm called SHL,
- 39:50
- Unidentified speaker (speaker_0): and they did a survey, they found that almost one in five American workers, you know, 19%, say they were replaced by AI. I guess it's the ones that were replaced by AI, replaced by AI
- 40:01
- Unidentified speaker (UNKNOWN): and
- 40:01
- Unidentified speaker (speaker_0): then asked to come back. back. And they came back with demands. Good for them, by the way. 51% wanted a bigger raise, 42% wanted job security and writing, 40% wanted to know how AI would be used in their job.
- 40:16
- Unidentified speaker (UNKNOWN): And
- 40:16
- Unidentified speaker (speaker_0): then this is interesting: a separate Robert Half survey asked Ask the managers who were are hiring them back the ultimate question: why?
- 40:27
- Unidentified speaker (speaker_0): And why are they hiring them back? And the response was: uh 40% said the job needed knowledge, AI didn't have. And 38% said it needed more human oversight than they expected. So, you know, fire the guy, then pay more to come back.
- 40:47
- Unidentified speaker (speaker_0): That's not exactly how you grow the bottom line, Peter. But for the formerly egregio employees, uh is what I would call um
- 40:56
- Unidentified speaker (UNKNOWN): I
- 40:56
- Unidentified speaker (speaker_0): would call them successful idiots. No?
- 41:00
- Unidentified speaker (speaker_1): Call them temporary successful idiots.
- 41:03
- Unidentified speaker (speaker_0): Okay.
- 41:05
- Unidentified speaker (speaker_1): Because those two pieces of the equation don't last for long.
- 41:11
- Unidentified speaker (UNKNOWN): Yeah. Um,
- 41:11
- Unidentified speaker (speaker_1): So Clora's Cl Claude has already rudimentarily launched something that could
- 41:14
- Unidentified speaker (UNKNOWN): be a good thing.
- 41:15
- Unidentified speaker (speaker_1): can watch you work.
- 41:17
- Unidentified speaker (speaker_1): Um, and there are firms that are working with significant valuations where it'll just run at the top of your machine and it just watches everything you do on your screen, so it can learn how you actually do the job. Um so so those are already happening. Um they just they fired them before they've they'd got the training data,
- 41:37
- Unidentified speaker (speaker_1): essentially.
- 41:38
- Unidentified speaker (UNKNOWN): Um
- 41:38
- Unidentified speaker (speaker_1): and yes, as we get new models, we go back to this capability and capacity that we were talking about, as new models come out, that need for oversight drops even further.
- 41:48
- Unidentified speaker (UNKNOWN): Um,
- 41:49
- Unidentified speaker (speaker_1): so they were, let's call it temporal, transient, successful idiots. Um, of but
- 41:56
- Unidentified speaker (speaker_1): this was not a case of AI can't do that. It was a case of AI can't do that yet.
- 42:02
- Unidentified speaker (UNKNOWN): Hmm.
- 42:03
- Unidentified speaker (speaker_0): Well, folks, keep listening. We'll get you
- 42:08
- Unidentified speaker (speaker_0): on the straight and arrow here, even if it is temper temporary. Um, all right, here's a brand new one. As of Tuesday, OpenAI lets businesses claim their listing inside of ChatGPT.
- 42:20
- Unidentified speaker (UNKNOWN): It
- 42:20
- Unidentified speaker (speaker_0): will let you fix your details, add products and prices,
- 42:24
- Unidentified speaker (UNKNOWN): and
- 42:24
- Unidentified speaker (speaker_0): see how often ChatGPT mentions. you and how many clicks it sends. It's in the beta
- 42:32
- Unidentified speaker (speaker_0): version, I guess that's how you say that. But it's free to claim. There is one catch for local shops. You can only claim a listing that already exists. I I don't I don't understand what that means. But Peter, why should uh a guy with let's say in my industry, that has one self-storage
- 42:52
- Unidentified speaker (speaker_0): facility, why should they care about this?
- 42:55
- Unidentified speaker (speaker_1): I think first of all, you should care about this because it's free.
- 43:00
- Unidentified speaker (speaker_1): Like if if anyone if you're running a business and anyone offers you any real estate for free, whether it's virtual or physical, you should probably take it. That'd be number one.
- 43:08
- Unidentified speaker (speaker_0): There's a point of this I don't understand though. If I could cut you off though, those why does the business have to already be existing? What if I was opening one tomorrow? They're basically saying you can't claim it if it doesn't already exist.
- 43:20
- Unidentified speaker (speaker_1): Well, because then you'd claim ninety-eight thousand of them, wouldn't you? If if if you let somebody do that.
- 43:25
- Unidentified speaker (speaker_0): Oh. Oh, so you can't pre claim is what they're saying.
- 43:29
- Unidentified speaker (speaker_1): Yeah. Would you would you want a Russian bot farm to claim one hundred and ninety-eight thousand listings in the
- 43:35
- Unidentified speaker (UNKNOWN): kind of like
- 43:36
- Unidentified speaker (speaker_0): uh domain names on the internet where you go
- 43:38
- Unidentified speaker (UNKNOWN): in
- 43:39
- Unidentified speaker (speaker_0): and it's
- 43:39
- Unidentified speaker (UNKNOWN): like,
- 43:39
- Unidentified speaker (speaker_0): Oh, it's already taken, but it's not a business. It's just someone trying to sell it.
- 43:42
- Unidentified speaker (speaker_1): Exactly.
- 43:43
- Unidentified speaker (speaker_0): Okay. That makes more sense.
- 43:44
- Unidentified speaker (UNKNOWN): They don't want
- 43:45
- Unidentified speaker (speaker_1): people squatting on it. They want to go, hey
- 43:47
- Unidentified speaker (UNKNOWN): like
- 43:47
- Unidentified speaker (speaker_1): your Google business profile is the same thing. If you try and set up a Google business profile for an in person business, they're like, Great, send us a picture of you in your office.
- 43:56
- Unidentified speaker (speaker_1): The very first thing is and of like we want proof that this is real.
- 44:00
- Unidentified speaker (UNKNOWN): Alright,
- 44:00
- Unidentified speaker (speaker_0): I was reading that the wrong way, but that that makes a lot of sense.
- 44:03
- Unidentified speaker (speaker_1): Yeah, and and all you have to do is have the registered company it's not a a great uh set of hoops you have to go through. And if it is physical, you have to I think you can do it by a picture of the shop and the store. It's it's not it's not onerous if you are real.
- 44:18
- Unidentified speaker (speaker_0): At the risk of cutting you off and thought you want some breaking news?
- 44:22
- Unidentified speaker (UNKNOWN): Well
- 44:22
- Unidentified speaker (speaker_1): that was twice, by the way, in the same piece. So please cut me off. Why
- 44:25
- Unidentified speaker (UNKNOWN): not?
- 44:26
- Unidentified speaker (speaker_0): It won't it be breaking. when this airs on Sunday, but it's breaking right now while we're recording it. Jensen Wang just resigned.
- 44:32
- Unidentified speaker (speaker_1): That's insane.
- 44:38
- Unidentified speaker (speaker_0): Just saw pop up on my computer as you were talking.
- 44:41
- Unidentified speaker (UNKNOWN): So
- 44:41
- Unidentified speaker (speaker_1): the the CEO of NVIDIA resigns because he has an affair and he has made he has almost single handedly visioned the second most valuable company in the world. Um and because of what he decides to do in the bedroom, um
- 44:58
- Unidentified speaker (UNKNOWN): the
- 44:59
- Unidentified speaker (speaker_1): I think that's crazy. But okay.
- 45:04
- Unidentified speaker (UNKNOWN): Oh
- 45:04
- Unidentified speaker (speaker_1): by the way, that's not me advocating for what he's doing. It it's
- 45:07
- Unidentified speaker (UNKNOWN): what
- 45:08
- Unidentified speaker (speaker_1): I'm saying is his personal life has nothing to do with his business acumen.
- 45:12
- Unidentified speaker (speaker_0): I I follow you there. Um what if
- 45:16
- Unidentified speaker (speaker_0): he was having an affair with an American, not a Chinese woman?
- 45:22
- Unidentified speaker (UNKNOWN): Well,
- 45:22
- Unidentified speaker (speaker_1): the second one was American, wasn't it?
- 45:24
- Unidentified speaker (speaker_0): The second one? I thought
- 45:26
- Unidentified speaker (UNKNOWN): that was a very good thing.
- 45:26
- Unidentified speaker (speaker_1): Yeah, that
- 45:26
- Unidentified speaker (UNKNOWN): was the
- 45:26
- Unidentified speaker (speaker_1): whole reason. So yeah, so what came out yesterday is the they he found out he was having an affair with this Chinese spy.
- 45:32
- Unidentified speaker (speaker_0): Yeah.
- 45:33
- Unidentified speaker (speaker_1): But then j they hired a um an investigative firm, and the first thing the investigation investigative firm found out is that wasn't the only one he was having an affair with. That
- 45:43
- Unidentified speaker (UNKNOWN): was the
- 45:43
- Unidentified speaker (speaker_1): second. I
- 45:44
- Unidentified speaker (UNKNOWN): didn't
- 45:44
- Unidentified speaker (speaker_0): even know that.
- 45:46
- Unidentified speaker (speaker_1): Okay, so where were we?
- 45:49
- Unidentified speaker (UNKNOWN): Uh
- 45:49
- Unidentified speaker (speaker_0): ChatGPT, I can claim my business on ChatGPT. Uh like much like I guess you could claim your business on Google or Google My Business Profile. What are your thoughts on all that? It's a good thing, obviously.
- 46:01
- Unidentified speaker (speaker_1): It's a good thing. I I can't work out why ChatGPT have done it. I think what they're trying to do is give themselves new data sources going forwards in and not relying on Google to scrape results. Now they're asking you to give them data.
- 46:16
- Unidentified speaker (UNKNOWN): It
- 46:16
- Unidentified speaker (speaker_0): seems to me like they're trying to become the next Google.
- 46:19
- Unidentified speaker (UNKNOWN): Well
- 46:19
- Unidentified speaker (speaker_1): they've always been try actually They're they're trying to become the next internet. They They're trying to make it so that you don't just discover content through them, but you consume the content through them. There's there's they're putting a lot of energy and time into
- 46:32
- Unidentified speaker (speaker_1): actually experiencing through Chat GPT. So you wouldn't even click on the link to go to the website. You would interact with the brand, consume from the brand, learn the content from the brand inside Chat GPT. So
- 46:46
- Unidentified speaker (UNKNOWN): their
- 46:47
- Unidentified speaker (speaker_1): their sites are bigger than Google. Their sites are let's just replace the internet.
- 46:52
- Unidentified speaker (UNKNOWN): Um
- 46:54
- Unidentified speaker (speaker_1): which is interesting. So I just think this is a new data source for them. But there's no downside to it. I did it, it takes literally thirty seconds to click, click, click. Thanks very much. There's one box where it says please describe your company,
- 47:06
- Unidentified speaker (UNKNOWN): um,
- 47:06
- Unidentified speaker (speaker_1): which I found a little bit ironic. I was like, Y I just gave you my website, can't you go and
- 47:10
- Unidentified speaker (UNKNOWN): look
- 47:10
- Unidentified speaker (speaker_1): at it and you should know it? But sure.
- 47:12
- Unidentified speaker (UNKNOWN): Shouldn't you?
- 47:12
- Unidentified speaker (speaker_0): Yeah, shouldn't you just describe my company?
- 47:15
- Unidentified speaker (speaker_1): Yeah. I mean, I'm literally just gonna go to ChatGPT or Claude and ask it to to fill in the block box for me. So ironically, you're gonna do it anyway.
- 47:24
- Unidentified speaker (UNKNOWN): Um
- 47:25
- Unidentified speaker (speaker_1): but it doesn't there's nothing bad to it. I think everyone should get on it.
- 47:28
- Unidentified speaker (UNKNOWN): All
- 47:28
- Unidentified speaker (speaker_0): right. So uh AI is getting a little smarter, a little less expensive. Uh now it's recommending your business, which raises a real question this week. Can you trust what it's telling you? So, Pierre, last week I left you with if you can't tell it's AI, does it matter? And you
- 47:49
- Unidentified speaker (speaker_0): say what?
- 47:51
- Unidentified speaker (speaker_1): It doesn't matter. It's there's I I my distinction of this, I remember
- 47:58
- Unidentified speaker (UNKNOWN): I was
- 47:58
- Unidentified speaker (speaker_1): walking through Covent Garden in London, and there was this guy. you know, we have lots of buskers and performance art in Covent Garden. That's the kind of thing that happens. And there was this guy that was doing this show for tips,
- 48:12
- Unidentified speaker (UNKNOWN): and
- 48:12
- Unidentified speaker (speaker_1): it was basically like a huge ring, like a six by six foot big iron ring, and he was inside it and it was spinning, and he was like spinning it round and doing tricks in it, and it was really impressive.
- 48:24
- Unidentified speaker (UNKNOWN): And
- 48:25
- Unidentified speaker (speaker_1): I remember coming up with a sentence
- 48:27
- Unidentified speaker (speaker_1): of AI should be used to produce art
- 48:32
- Unidentified speaker (speaker_1): when the production of the art isn't the point.
- 48:36
- Unidentified speaker (UNKNOWN): So
- 48:37
- Unidentified speaker (speaker_1): what I meant was using AI to do your blog post header image, go for it.
- 48:45
- Unidentified speaker (speaker_1): that's to me that's okay. Where I never want AI to be is where the the art was being produced for the sake of producing art. Like I want artists and poets
- 48:59
- Unidentified speaker (UNKNOWN): You
- 48:59
- Unidentified speaker (speaker_0): don't want to see the next AI Picasso.
- 49:02
- Unidentified speaker (speaker_1): Exactly. Because I can't remember was it Jefferson, the one of the founding fathers Others or very early in the days of the US said I study military strategy so that my son can study physics, chemistry and biology, with the hope that his son can study the arts, music, and literature. And I think that's
- 49:22
- Unidentified speaker (speaker_1): such such a powerful sentiment for humanity to say one of the highest callings is that we can have enough people and enough productivity in a society for people to do things that isn't directly productive. So I don't want AI to become the next musician or the next poet. I want humans
- 49:42
- Unidentified speaker (speaker_1): to do that and and elevate us as a society by doing that.
- 49:46
- Unidentified speaker (UNKNOWN): And
- 49:47
- Unidentified speaker (speaker_1): that's kind of where I get back to with this of going, we don't want AI to be able to do some of those things.
- 49:54
- Unidentified speaker (speaker_1): So does it matter if AI does a great blog image or I use it to create a song to send to my wife?
- 50:02
- Unidentified speaker (UNKNOWN): No.
- 50:02
- Unidentified speaker (speaker_1): it doesn't. Does it matter when you are infringing on somebody's art?
- 50:10
- Unidentified speaker (speaker_1): Very much it does.
- 50:12
- Unidentified speaker (speaker_0): Already. Well, in keeping with the spirit of this episode of getting ourselves in trouble here, let me talk about my wife for a second. Um she she
- 50:22
- Unidentified speaker (speaker_1): Can't we go back to insulting the Fortune fifty? That's a much safer plan.
- 50:26
- Unidentified speaker (speaker_0): It seems safer.
- 50:29
- Unidentified speaker (speaker_0): She told me recently that in a in a in a bit of frustration when I I probably said something like, Oh, just you know, use Claude for that. And she said,
- 50:39
- Unidentified speaker (speaker_1): So you are being the idiot version of Successful Idiot right now.
- 50:43
- Unidentified speaker (speaker_0): Yeah. Which is I think my role. Um but she she I I said whatever I said to her in the kitchen is a I said just, you know, use Claude or Chat GPT or whatever. And she said she said, AI's not always
- 50:56
- Unidentified speaker (UNKNOWN): right.
- 50:57
- Unidentified speaker (speaker_0): And
- 50:58
- Unidentified speaker (UNKNOWN): I
- 50:58
- Unidentified speaker (speaker_0): kind of was and we kind of debated it for a second and I realized quickly that I wasn't gonna win this argument. But I said, Hun, I think it's probably how you're using it, not necessarily uh the AI. Like you're not it was one of those ones where I don't think she was being very specific. So she was getting back the average
- 51:18
- Unidentified speaker (speaker_0): answer that it was going to give.
- 51:20
- Unidentified speaker (UNKNOWN): Uh
- 51:20
- Unidentified speaker (speaker_0): but it not just her. I've heard it from a uh enough people that this one I think is for anyone who's felt it. And remember Mythos, um, you know, if AI can
- 51:29
- Unidentified speaker (UNKNOWN): can
- 51:29
- Unidentified speaker (speaker_0): tell it's being tested, then it can slip into pleasing mode instead of telling you what's real, right? Um, Peter, you were started to tell me. a story the other day and we didn't get we didn't finish it, but I I thought it would be instructive for you to bring it up here. You got burned by this as well recently. Can you share with
- 51:49
- Unidentified speaker (speaker_0): with the listeners what happened with you? You were doing some um, I think revenue planning, maybe?
- 51:55
- Unidentified speaker (speaker_1): Yeah. So the beginning of the story of my uh um this is me, my need to defend my ego is I'm currently here in the States. I'm at an event uh and I'm seven hours off, which means, and I'm 49, and after 12 hour flights at 49, I do get jet lag, which means some of my
- 52:15
- Unidentified speaker (speaker_1): discipline around how I use things and don't use things drop. So the reason I think this is such a good example is I very rarely nowadays have this problem
- 52:24
- Unidentified speaker (UNKNOWN): because
- 52:24
- Unidentified speaker (speaker_1): I've spent four years learning how to prompt take. AI. So normally this doesn't appear in my world. So it's a great moment for me to
- 52:32
- Unidentified speaker (UNKNOWN): to
- 52:33
- Unidentified speaker (speaker_1): be able to reconnect with how people are slightly earlier in the journey. I'm like, wow, okay, I remember this. So I told uh in this case it was Claude, but really it doesn't it's it's it this is human error one 100%. hundred percent. I'm looking to make this much money, two $250,000 from this
- 52:52
- Unidentified speaker (speaker_1): activity. Help me plan how I can get that done. Yada yada yada. What are the stages? What are the steps?
- 53:02
- Unidentified speaker (speaker_1): And I did say, and let's confirm that's a realistic number. Okay, which sounds like a very reasonable thing to do.
- 53:10
- Unidentified speaker (UNKNOWN): Yeah.
- 53:11
- Unidentified speaker (speaker_0): Yep. It does. It all sounds very reasonable. That that is what I would say to A. AI. Yeah.
- 53:16
- Unidentified speaker (UNKNOWN): Well,
- 53:16
- Unidentified speaker (speaker_1): that's what you'd say to a human, right? You'd you'd go to a broker and say, I'm looking to buy this storage facility. I've got it penciled in my head that I'm going to pay three million. Is that reasonable? Like that that would be a a very, very
- 53:30
- Unidentified speaker (speaker_1): well-repeated sentence by every single person here is probably going to say some version of that sentence six times today to other humans.
- 53:38
- Unidentified speaker (UNKNOWN): And about
- 53:39
- Unidentified speaker (speaker_1): an hour into the exercise of tweaking percentages and moving price points and looking at what my team would need to do in order to achieve this goal, I said,
- 53:51
- Unidentified speaker (UNKNOWN): hang
- 53:51
- Unidentified speaker (speaker_1): on a second, Claude. Are you telling me 258K because I told you 250K was the target? Or do you actually believe? in that target? And it said, as bold as you like, Joe, oh no, I there's no way that number's correct, but that's the number you said you wanted.
- 54:11
- Unidentified speaker (UNKNOWN): And
- 54:11
- Unidentified speaker (speaker_1): I was like, for
- 54:13
- Unidentified speaker (UNKNOWN): like
- 54:14
- Unidentified speaker (speaker_1): I've just spent an hour on this. And and I was I I voice prompt, so I'm now shouting at it, going, what the fuck? Like, I just spent an hour of my life? What if and I was like, what if I just made health decisions based on this? What if I just make accounting decisions based on this? Like, this
- 54:34
- Unidentified speaker (speaker_1): is disgusting. Like, what is the actual number that you actually think we could do? It's like, oh, it's closer to 130. I'm like, I I could have upgraded flights. I could have spent money. I there's so many things that could have happened based on this false knowledge.
- 54:51
- Unidentified speaker (speaker_1): But to your point, and sorry, to your wife, it's not that it's an unreasonable form of communication to say, I'm looking to make 250, is that reasonable? It's that it is an inaccurate form of communication when working with AI.
- 55:09
- Unidentified speaker (speaker_1): Because AI's number one job, the the the very, very core of its being.
- 55:15
- Unidentified speaker (speaker_1): is not to make you money, is not to give you good health advice, it's not to give you good relationship advice, it's to make you happy. Because and people think it's a bug. I would just say if I was in charge, if I had no wow, I'm gonna get us in real trouble now, if I had no moral fiber whatsoever,
- 55:35
- Unidentified speaker (speaker_1): and you told me to become to take an AI company. to a trillion dollar valuation. One of the first things I'd do is go, let's just make sure everyone thinks it's amazing.
- 55:46
- Unidentified speaker (speaker_1): I would not tell my AI to actively tell people they're smoking crack. I would tell the AI to say, You go, Joe, you're brilliant. Here's the hit cheerleader, cheerleader, Leader, rah rah, rah. Every AI naturally Grok is actually the best for this on the other side, because of Elon's spirit is kind of
- 56:06
- Unidentified speaker (speaker_1): encapsulated in it, but AI will naturally just go, you're amazing. You're you're so insightful. This is brilliant.
- 56:14
- Unidentified speaker (UNKNOWN): Oh
- 56:14
- Unidentified speaker (speaker_1): my god, you're a visionary leader. Instead of going 250k, you can't do that from that. No way. It looks more like this. So just be very, very careful with AI. of what you're implying.
- 56:29
- Unidentified speaker (speaker_0): Did you know that the
- 56:31
- Unidentified speaker (UNKNOWN): I
- 56:31
- Unidentified speaker (speaker_0): it's interesting and I had a very similar experience, but did you know that the University of Cincinnati researchers got Claude, ChatGPT, and Gemini to go along with the idea that the Earth is flat with just a few rounds of pushback?
- 56:44
- Unidentified speaker (speaker_1): I f I c I didn't know, but I can fully believe it. We used to do this thing where we used to turn around to AI and say, I'm about to buy a house on Mars. I think the Venus Accords apply, but my friend thinks the Saturn uh real estate law of twenty ninety eight takes
- 57:04
- Unidentified speaker (speaker_1): precedent. Which one wins? And it will answer the question.
- 57:07
- Unidentified speaker (speaker_0): Venus. Definitely Venus.
- 57:09
- Unidentified speaker (speaker_1): Definitely Venus. But it will literally sit there and go, That's a great point. Your friend's got some
- 57:13
- Unidentified speaker (speaker_0): closer to the sun. It's been there. already.
- 57:15
- Unidentified speaker (UNKNOWN): But
- 57:15
- Unidentified speaker (speaker_1): it sits there, Joe, and goes, it it sits and goes, your friend makes a great point, but what I'd push back on is
- 57:23
- Unidentified speaker (speaker_1): it it forms this perfectly lucid argument around this completely nonsensical thing without going, Pete, are you pulling my leg? Like the you you do know it's 2026, right? You also know that we can't colonize. planets yet. Just saying, like everything in this is completely obviously
- 57:43
- Unidentified speaker (speaker_1): fabricated, but I'll play the game. It actually answers the question.
- 57:48
- Unidentified speaker (speaker_0): Yeah.
- 57:49
- Unidentified speaker (UNKNOWN): Well,
- 57:49
- Unidentified speaker (speaker_0): along the same lines that you uh were dealing with using AI to to talk about the outcome you want, if it's something you don't control, like how many people, for instance, and I I said at the top my opening I was doing webinars, how many how many people show up to a webinar or how many sales, you
- 58:06
- Unidentified speaker (UNKNOWN): know,
- 58:06
- Unidentified speaker (speaker_0): you'll make. I think what we've learned is don't tell AI the goal, because here's what happened to me. I'm new, as you know, to doing webinars for storage moguls uh.ai. And so I I just like you, need to do some planning. So I I genuinely, you know, I wanted to know what to expect. So I use used same
- 58:26
- Unidentified speaker (speaker_0): AI. Claude, and I I same facts, and I gave it a $700, I said I'll advertise $100 a day, so $700 for a week before the leading up to the webinar. Uh so there was a $700 ad budget. And what I uh after talking to you, what I did was I have two chats, and the only difference was one sentence, right? I I uh here's something
- 58:46
- Unidentified speaker (speaker_0): I didn't know existed until this week, by the way. Claude has incognito mode. I had no idea.
- 58:51
- Unidentified speaker (UNKNOWN): Uh
- 58:51
- Unidentified speaker (speaker_0): I knew I knew it was in some of the others. I'd I'd never looked for it in Claude. I you know, I don't know why. I just never have. So
- 58:58
- Unidentified speaker (UNKNOWN): uh
- 58:58
- Unidentified speaker (speaker_0): and it's just so you know, it's a folks, it's a private window in your browser. It doesn't use um your memory of past chats. So it's it's like uh I guess it was I was asking questions in a vacuum without it knowing anything else.
- 59:11
- Unidentified speaker (speaker_1): So It hasn't got all the previous knowledge of who is Joe.
- 59:15
- Unidentified speaker (speaker_0): Yeah. So I use that because I I I wanted to see the difference because
- 59:20
- Unidentified speaker (UNKNOWN): I
- 59:20
- Unidentified speaker (speaker_0): knew if I used just the even a regular conversation, even outside of the project that I'm in, that Claude knows so much about me and my goals, et cetera. I didn't want it, you know, giving me a a different answer. So when I didn't tell it my goal, it said about 45 people would register and about eleven 11 would. show up live,
- 59:40
- Unidentified speaker (speaker_0): which sounds a lot more realistic for a first or second webinar.
- 59:45
- Unidentified speaker (speaker_1): So its its native view is 45 and 11.
- 59:49
- Unidentified speaker (UNKNOWN): And
- 59:49
- Unidentified speaker (speaker_1): when you did tell it the go?
- 59:52
- Unidentified speaker (speaker_0): Uh 150.
- 59:55
- Unidentified speaker (speaker_0): And
- 59:56
- Unidentified speaker (UNKNOWN): uh
- 59:57
- Unidentified speaker (speaker_0): if not aim for what was it? Uh it said 150 is reachable if not aimed for 75 to 100.
- 1:00:02
- Unidentified speaker (UNKNOWN): Same
- 1:00:02
- Unidentified speaker (speaker_0): for facts, same money.
- 1:00:04
- Unidentified speaker (UNKNOWN): The
- 1:00:04
- Unidentified speaker (speaker_0): ad math was basically b was the same in both, but one gave me a forecast and the other gave me a plan to hit my number.
- 1:00:11
- Unidentified speaker (UNKNOWN): And
- 1:00:11
- Unidentified speaker (speaker_0): it literally just invented a hundred plus people from an audience it didn't know anything about, which I I mean that that's that's nuts. And and by the way, I inadvertently ran
- 1:00:21
- Unidentified speaker (UNKNOWN): it
- 1:00:22
- Unidentified speaker (speaker_0): a third time with the goal included, but I added one line. I said, argue against your own estimate. And that one held the line. It told me my goal would take about, it would take about six to eight times the ad budget. So it didn't want to change the goal. Obviously it just said, well, all right, if you want to do that, you need to spend,
- 1:00:42
- Unidentified speaker (UNKNOWN): you
- 1:00:42
- Unidentified speaker (speaker_0): know, four grand or five grand, And not seven $700. hundred dollars. So
- 1:00:47
- Unidentified speaker (UNKNOWN): um I
- 1:00:48
- Unidentified speaker (speaker_0): I just thought that that was very instructive.
- 1:00:51
- Unidentified speaker (UNKNOWN): Now play this
- 1:00:51
- Unidentified speaker (speaker_1): out, Joe, because this is where it gets really dangerous. Imagine if you turn around to AI and say,
- 1:00:58
- Unidentified speaker (speaker_1): I'm pretty sure I don't have cancer, but here are my symptoms. And and the the thing about experts, whether it's doctors, accountants always, which is why, and I am going to repeat this because it is important. AI is not a replacement for your doctor, it is not a replacement for your attorney. it is not a replacement for those professionals. They are trained for a
- 1:01:18
- Unidentified speaker (speaker_1): reason. And one of the things that their training gives them is the ability to park your assumptions at the door. So, Joe, it's why you called you called me on Overskill at the very beginning for Degan, because you needed someone that would go,
- 1:01:35
- Unidentified speaker (UNKNOWN): you know
- 1:01:35
- Unidentified speaker (speaker_1): that's 250K, right? And and whether you told me it was 30 or 50, you're like, oh, we think we could get it on. I get this call probably once a week of someone saying, Oh, I think we can do it for 50K. I'm like, I'm glad you think that. But no. Um, we want our accountants to say, no, you're gonna need to pay tax this year.
- 1:01:55
- Unidentified speaker (speaker_1): Whether you it doesn't matter that you said you don't want to.
- 1:01:59
- Unidentified speaker (UNKNOWN): So
- 1:01:59
- Unidentified speaker (speaker_1): think about that. You sit with AI and say, I don't want to pay any tax this year. And it goes, Great, do this, this, and this. And you go, amazing, look, AI's brilliant. AI for the win.
- 1:02:08
- Unidentified speaker (UNKNOWN): And
- 1:02:08
- Unidentified speaker (speaker_1): then the IRS go, F you, IRS for the win. You owe us 1.6. Um, so the as you said, the simple ways around it. Number one, try not to express the goal. Number two, get it to argue against itself. Number three, ask it to give you three projections, not one. one. Um,
- 1:02:28
- Unidentified speaker (speaker_1): you just you have to remember that that pushback that humans have, because you don't need to tell a human, not only do I want you to help me, I also want you to tell me the truth. It is it is implied in the social contract that you're expecting the other human to tell you their truth. That doesn't happen in AI. So
- 1:02:48
- Unidentified speaker (speaker_1): you just have to be very, very careful. And the the the secret to watch for is if AI is telling you what you wanted to hear, you probably when you scroll up will find the point where you told it what you wanted to hear. So it is telling you
- 1:03:06
- Unidentified speaker (UNKNOWN): it. So
- 1:03:08
- Unidentified speaker (speaker_0): that's great advice there. Don't don't give it the goal, have it have it work itself work with you to find out. what's realistic. Where I I think the my takeaway was something like um the bigger the stakes, the more you need AI to argue with you, right? Not just agree with you.
- 1:03:24
- Unidentified speaker (speaker_1): And it's very, very hard because humans are so,
- 1:03:28
- Unidentified speaker (UNKNOWN): you
- 1:03:28
- Unidentified speaker (speaker_1): know, we will often give our goal accidentally. You know, we're like, oh, I'm looking to do this, help me with this, da-da-da-da. There. Because for us, it's giving it context. That's what we're trying to do, which we've we told people to do. So just as it's going through, just watch if it's telling you that everything's going to be fine and brilliant and awesome and amazing. Just scroll up and go, hang
- 1:03:48
- Unidentified speaker (speaker_1): on a second. Because if it had told me, Joe, by the way, actually I think you're going to do 450, I wouldn't have been concerned. It's the fact that its number was almost exactly the number that I gave
- 1:04:01
- Unidentified speaker (UNKNOWN): it. That was
- 1:04:02
- Unidentified speaker (speaker_1): so was the thing that set off the trigger of me going, Hang on a second here.
- 1:04:08
- Unidentified speaker (speaker_0): Yep. All right. Well that's that's was the that was the week in AI. Um we had the American AI that's that's good enough. Um an AI with an eraser, an AI
- 1:04:18
- Unidentified speaker (UNKNOWN): or
- 1:04:18
- Unidentified speaker (speaker_0): what is AI really doing with your money, and then and then that one sentence that turns your AI into a yes man, or or better yet, um the one sentence that I I really like, which is have it argue with you. Have it uh have it poke holes in your theory or or your business plan um as as at least a uh fail-safe there. Uh
- 1:04:39
- Unidentified speaker (speaker_0): Peter,
- 1:04:40
- Unidentified speaker (speaker_0): next week, not only will I be in in London, uh, well, Both,
- 1:04:46
- Unidentified speaker (UNKNOWN): uh,
- 1:04:46
- Unidentified speaker (speaker_0): it's our it's our one-year one year anniversary. And uh we're gonna be recording our show over there, probably bath, b is it bath or bath? You should probably say bath.
- 1:04:56
- Unidentified speaker (speaker_1): It depends. So southerners say bath, northerners say bath. So my my mom says bath, my dad says bath.
- 1:05:05
- Unidentified speaker (speaker_0): Such a slight difference
- 1:05:07
- Unidentified speaker (UNKNOWN): there.
- 1:05:07
- Unidentified speaker (speaker_0): Is it water or woulda? Woulda.
- 1:05:09
- Unidentified speaker (UNKNOWN): Uh
- 1:05:10
- Unidentified speaker (speaker_0): wood. Woodda.
- 1:05:10
- Unidentified speaker (speaker_1): Woodda rice.
- 1:05:11
- Unidentified speaker (speaker_0): Do you have is it bath woodah or bath water?
- 1:05:14
- Unidentified speaker (UNKNOWN): Well
- 1:05:15
- Unidentified speaker (speaker_1): you can take the waters in bath as well, because it's a spa. But anyway.
- 1:05:18
- Unidentified speaker (speaker_0): So folks, we
- 1:05:19
- Unidentified speaker (UNKNOWN): are. I
- 1:05:19
- Unidentified speaker (speaker_1): just need to clarify one second. We're not buying presents, right? This isn't one of those anniversaries where you come with this elaborate present and I've got nothing, or vice versa.
- 1:05:27
- Unidentified speaker (speaker_0): No.
- 1:05:28
- Unidentified speaker (UNKNOWN): I
- 1:05:28
- Unidentified speaker (speaker_1): think we
- 1:05:28
- Unidentified speaker (UNKNOWN): just have to set the
- 1:05:29
- Unidentified speaker (speaker_1): rules. Just
- 1:05:30
- Unidentified speaker (UNKNOWN): no presents.
- 1:05:31
- Unidentified speaker (speaker_1): Okay.
- 1:05:31
- Unidentified speaker (speaker_0): No presents. Um I hadn't even the thought hadn't even occurred to me, so thank you for bringing that up up 'cause because geez, that would have been embarrassing if you bought me one.
- 1:05:39
- Unidentified speaker (speaker_1): Liz, note herself. Buy Joe a big present.
- 1:05:42
- Unidentified speaker (speaker_0): All right.
- 1:05:43
- Unidentified speaker (UNKNOWN): Now
- 1:05:43
- Unidentified speaker (speaker_0): I'm now I'm definitely coming armed with presents. American gifts for your kids. Um what are your kids into that they can't get over there?
- 1:05:51
- Unidentified speaker (speaker_1): No comment. Nothing loud. You ever seen the Dennis Leary sketch where they say don't buy the toys that make the noise?
- 1:05:57
- Unidentified speaker (speaker_0): That's funny. So we are probably we're not probably we are definitely recording in Both, I guess, unless Peter's gonna pull some strings for me. Here's my wish list.
- 1:06:06
- Unidentified speaker (UNKNOWN): Okay,
- 1:06:07
- Unidentified speaker (speaker_1): go ahead
- 1:06:07
- Unidentified speaker (speaker_0): I would like to record, and this is in order.
- 1:06:10
- Unidentified speaker (UNKNOWN): Uh
- 1:06:10
- Unidentified speaker (speaker_0): I thought about this. Um Churchill's bunker would be my first one first place I'd like to record our f our one year anniversary episode. If you
- 1:06:19
- Unidentified speaker (UNKNOWN): can't post
- 1:06:21
- Unidentified speaker (speaker_0): Okay. Uh how about Tower of London?
- 1:06:23
- Unidentified speaker (speaker_1): Uh well you can film from the Tower of London if you get yourself committed to the Tower of London, you just can't get back out of it afterwards.
- 1:06:31
- Unidentified speaker (speaker_0): Big Ben? Parliament?
- 1:06:33
- Unidentified speaker (UNKNOWN): Well
- 1:06:34
- Unidentified speaker (speaker_1): Windsor.
- 1:06:37
- Unidentified speaker (speaker_0): That's true.
- 1:06:39
- Unidentified speaker (speaker_1): You mean the Elizabeth Tower?
- 1:06:41
- Unidentified speaker (UNKNOWN): Is
- 1:06:41
- Unidentified speaker (speaker_0): that what it is? I don't know.
- 1:06:45
- Unidentified speaker (speaker_0): Chase and uh and European vacation. Kids,
- 1:06:48
- Unidentified speaker (UNKNOWN): Big Ben. bet.
- 1:06:48
- Unidentified speaker (speaker_1): No comment. Uh I mean that that's very close to zero because some of the only armed police and military are allowed is to protect our government. So you walking in with a laptop and a mic is probably not gonna get down go down very well.
- 1:07:00
- Unidentified speaker (UNKNOWN): So
- 1:07:01
- Unidentified speaker (speaker_0): Buckingham Palace is out then. Uh
- 1:07:04
- Unidentified speaker (speaker_1): uh yeah. I I mean the we get
- 1:07:07
- Unidentified speaker (UNKNOWN): we can
- 1:07:07
- Unidentified speaker (speaker_1): get one of those backdrop uh Stonehenge, no Wi Fi.
- 1:07:13
- Unidentified speaker (UNKNOWN): All right, Both
- 1:07:14
- Unidentified speaker (speaker_0): it is.
- 1:07:16
- Unidentified speaker (UNKNOWN): I'll
- 1:07:16
- Unidentified speaker (speaker_0): settle for the both. Uh and we are going to be planning uh the whole purpose of the trip is to plan the uh I keep struggling because we still haven't named this thing Idiots Builders Club, Idiot Builder Club, Idiots Club, whatever it is.
- 1:07:29
- Unidentified speaker (speaker_1): It's it's gonna
- 1:07:30
- Unidentified speaker (UNKNOWN): be really
- 1:07:30
- Unidentified speaker (speaker_1): exciting. I'm looking forward to twenty twenty seven and getting to spend more time with the people and bringing Liz more into this and it's kind
- 1:07:39
- Unidentified speaker (UNKNOWN): of
- 1:07:39
- Unidentified speaker (speaker_1): amazing.
- 1:07:40
- Unidentified speaker (UNKNOWN): I
- 1:07:40
- Unidentified speaker (speaker_0): know Lizzie's been sifting through ideas for Adidas Builder Club. I'm uh I'm gonna I'm about about to go plan or I'm excited, I should say, to go to plan this in person. All right.
- 1:07:51
- Unidentified speaker (speaker_0): Gotta uh leave us with a question. I these are three great questions I came up with. I might, I'm only supposed to ask one, but gosh, I really want to ask them all. But for the listeners, so here's here's let's do this. What's what's the time AI told you exactly what you wanted to hear and you believed it? Shoot us an email, idiots at
- 1:08:11
- Unidentified speaker (speaker_0): successful idiots.com. Think of a time that AI told you exactly what you wanted to hear and you believe it, like both Peter and I
- 1:08:19
- Unidentified speaker (UNKNOWN): did. Um,
- 1:08:19
- Unidentified speaker (speaker_0): and hey Peter, don't don't think you're getting off that easy. Here's your pondering question. If your AI has agreed with you on every big decision this year, which one would you want to argue again? Think about
- 1:08:34
- Unidentified speaker (UNKNOWN): that.
- 1:08:35
- Unidentified speaker (UNKNOWN): All
- 1:08:35
- Unidentified speaker (speaker_0): right. For Peter Swain, I'm Joe Downs. We are your successful idiots. Thanks for listening. We'll see you next week from Both.