Kevin Schoovaerts details an AI pipeline for microcap investing using Claude, emphasizing human oversight and data quality.
Summary
Kevin Schoovaerts, founder of Atomical Capital, outlines a human-in-the-loop AI investment pipeline using four agents: Hunter, Researcher, Contrarian, and Tracker. He argues that data quality is a prerequisite for AI success and warns against trusting LLM outputs without verification, likening the AI to a junior analyst and the human to a senior judge. Schoovaerts advocates for using Claude Skills and Python to add determinism to statistical models and emphasizes the importance of a 'vault' of past decisions to guide the AI. He demonstrates how AI can flag errors in past theses and analyze portfolio concentrations via Interactive Brokers, while stressing that investors must understand fundamentals to evaluate AI output.
Episode sections
- 0:00 – 2:57
AI Pipeline Overview
Speaker_0 describes a human-in-the-loop AI pipeline using four agents: Hunter (idea generation), Researcher (report creation), Contrarian (thesis pushback), and Tracker (portfolio monitoring). A central 'vault' stores the user's expertise and decisions to guide the agents, ensuring the human remains the final decision-maker.
- 2:57 – 4:43
Introduction and Background
Kevin Schoovaerts introduces himself as the founder of Atomical Capital, a nuclear physicist with 17 years of investing experience. He runs two newsletters: '100 Bagger Hunting' for microcap deep dives and 'Invest with AI' for teaching AI usage, specifically with Claude.
- 4:43 – 6:45
Feynman's Approach to Investing
Speaker_0 cites Richard Feynman as his favorite investor, emphasizing Feynman's teaching style and his principle to 'distrust experts and even distrust yourself.' He applies this by building investment theses and actively seeking evidence to destroy them, rather than confirm them.
- 6:45 – 8:48
Stairway Framework and AI History
The presentation follows a five-step 'stairway' framework: prompting, Claude Skills, data, agents, and memory. Speaker_0 notes that while ChatGPT reached 100 million users in two months, early equity analysis tools produced 'beautifully written nonsense' due to model limitations.
- 8:48 – 10:50
LLM as Junior, Human as Senior
Using a car analogy, Speaker_0 states the LLM is the engine and the human is the steering wheel. He argues users must maintain the mindset that the LLM is a junior preparing work for a senior human judge, who must never accept results at 'point blank range' without verification.
- 10:50 – 14:00
AI Adoption and Data Quality
Speaker_0 claims most individual investors rarely use AI for investing, citing a survey where 40% never use it. He contrasts this with a large asset management firm that spent two years cleaning data before building automation layers, emphasizing that data quality is a prerequisite for AI.
- 14:00 – 15:48
Three Components of AI Tools
The three components of an AI tool are data, prompt, and model (brain). Speaker_0 warns that bad data with a smart brain and good prompt is the most dangerous combination, as it produces super-confident but incorrect replies, whereas great data can compensate for vague prompts.
- 15:48 – 32:14
Claude Skills and Python
Skills are reusable software pieces that combine prompts with Python scripts to add determinism to the statistical nature of LLMs. Speaker_0 highlights that Claude can execute Python, allowing for reliable calculations and structured outputs that pure text prompts cannot guarantee.
- 32:14 – 32:51
Thesis Quality and Exit Rules
Speaker_0 emphasizes that the quality of the investment thesis is the most critical factor for an AI system to make judgments. He advises including specific exit rules in the thesis so the AI can evaluate when to sell.
- 32:51 – 34:33
Context, Memory, and the 'Genie' Analogy
Speaker_0 uses the Aladdin genie analogy to explain that LLMs must be kept 'in the lamp' by providing specific context and data rather than relying on general training. He suggests feeding the AI past investment decisions to create a feedback loop for better future judgments.
- 34:33 – 35:09
Building a 'Second Brain' and Junior Analyst
Speaker_0 describes setting up a local system where an AI reads and writes to local folders, acting as a 'junior analyst' that pushes back against the investor's thesis. He cites Andre Karpathy's approach to structuring knowledge as a model for this setup.
- 35:09 – 39:21
Automated Earnings Updates and Error Detection
Speaker_0 shares an anecdote where an AI agent analyzing a recent earnings update flagged a mistake from a year prior by comparing current data against past statements and counter-arguments, demonstrating the value of a robust data pipeline.
- 39:21 – 40:55
Sourcing Data for Thesis Trackers
Speaker_3 asks how to source information for a thesis tracker. Speaker_0 explains that US filings can be pulled via the SEC EDGAR system using Claude's API, while international data can come from APIs like Japan's EDNet or paid providers like EODHD.
- 40:55 – 1:02:28
Data Providers and Connectors
Speaker_0 mentions using EODHD for global microcap data due to its breadth, noting that Claude has thousands of connectors. He also lists Financial Modeling Prep and Fiscal.ai as alternatives, emphasizing the need to supplement data providers with direct website checks for materials like investor presentations.
Claims worth verifying
Speaker_0's AI pipeline consists of four agents: Hunter, Researcher, Contrarian, and Tracker, all using a central 'vault' of expertise.
Unidentified speaker (speaker_0) ·0:00
How to verify: Verify by reviewing the slide described at 0:00-0:45 or the 'Invest with AI' course materials/course website. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Richard Feynman won the Nobel Prize for Physics and is cited by Speaker_0 as his favorite investor for his teaching ability and distrust of experts.
Unidentified speaker (speaker_0) ·5:02
How to verify: Verify Feynman's Nobel Prize via Nobel Prize official website; verify Speaker_0's statement via transcript at 5:02. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
ChatGPT reached 100 million users in two months.
Unidentified speaker (speaker_0) ·7:41
How to verify: Verify via OpenAI press releases or major news reports from early 2026 regarding ChatGPT user milestones. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Facebook took 54 months to reach 100 million users.
Unidentified speaker (UNKNOWN) ·7:58
How to verify: Verify via Facebook/Meta historical user growth reports or tech industry analyses of Facebook's growth timeline. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
A survey of Speaker_0's audience showed 40% never use AI for investing and 37% use it occasionally.
Unidentified speaker (UNKNOWN) ·11:08
How to verify: Verify by checking Speaker_0's newsletter archives or social media posts from around the time of the presentation (October 2026) referencing this survey. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
A large asset management firm with 2,000 employees spent two years cleaning data before implementing AI.
Unidentified speaker (speaker_0) ·12:10
How to verify: Verify via the transcript at 12:10; cross-reference with public case studies of AI implementation in asset management if the firm is named (COO mentioned but not named). For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Claude Skills can include Python scripts to add determinism to LLM outputs.
Unidentified speaker (speaker_0) ·18:17
How to verify: Verify via Anthropic's documentation for Claude Code or Claude Skills, or Speaker_0's 'Invest with AI' course materials. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Anthropic created a free 'initiating coverage' skill in collaboration with JP Morgan.
Unidentified speaker (speaker_0) ·21:09
How to verify: Verify via Anthropic's official blog or Claude Skills marketplace for a skill developed with JP Morgan. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 analyzed 180 European banks using an AI agent in a specific example.
Unidentified speaker (UNKNOWN) ·26:15
How to verify: Verify via Speaker_0's newsletter '100 Bagger Hunting' or 'Invest with AI' content from around October 2026 detailing this analysis. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 held a position in 'Sunavave' (likely Sunovion or similar) where his AI tracker flagged a broken thesis in November 2025.
Unidentified speaker (speaker_0) ·30:24
How to verify: Verify via Speaker_0's portfolio updates or newsletter posts from late 2025/early 2026 discussing this position and the AI alert. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 states that if the investment thesis is vague, the LLM cannot judge effectively.
Unidentified speaker (speaker_0) ·32:14
How to verify: Listen to transcript at 32:14; verify by testing an LLM with vague vs. specific thesis prompts. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 claims he experienced an AI agent flagging a mistake he made about a year earlier during an earnings update by quoting his past counter-arguments.
Unidentified speaker (speaker_0) ·35:44
How to verify: Listen to transcript at 35:44; verify by checking Speaker_0's public writings or newsletter for the specific earnings update and error mentioned. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 states that Claude has an API to access the SEC EDGAR system.
Unidentified speaker (speaker_0) ·39:50
How to verify: Listen to transcript at 39:50; verify by checking Claude's documentation or API reference for SEC EDGAR integration. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Primary sourceSpeaker_0 uses EODHD as his data provider for global microcap data.
Unidentified speaker (speaker_0) ·40:39
How to verify: Listen to transcript at 40:39; verify by checking EODHD's website for API/MCP availability and Speaker_0's course materials. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 tested 'Opus five point five' and found it could create a DCF valuation in one go with no errors.
Unidentified speaker (speaker_0) ·53:15
How to verify: Listen to transcript at 53:15; verify by checking Anthropic's model release notes for 'Opus' versions or Speaker_0's course content for the specific model tested. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 states that Interactive Brokers has an MCP connection to Claude.
Unidentified speaker (speaker_3) ·43:21
How to verify: Listen to transcript at 43:21; verify by checking Interactive Brokers' API documentation or Claude's connector list for 'Interactive Brokers'. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 mentions that Japan has an API called EDNet for pulling filings.
Unidentified speaker (UNKNOWN) ·40:00
How to verify: Listen to transcript at 40:00; verify by searching for 'Japan EDNet API' or checking Japanese financial data provider documentation. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Primary sourceSpeaker_0 claims there are about 1500 or 2000 applications that have provided a connector to Claude.
Unidentified speaker (speaker_0) ·41:23
How to verify: Listen to transcript at 41:23; verify by checking Claude's official documentation or marketplace for the number of available connectors. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 states that he deactivated the setting in Claude that allows it to learn from his knowledge.
Unidentified speaker (speaker_0) ·1:01:53
How to verify: Listen to transcript at 1:01:53; verify by checking Claude's settings documentation regarding 'learning from user data'. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Speaker_0 recommends using AI to learn faster rather than as a shortcut for those new to investing.
Unidentified speaker (speaker_0) ·51:23
How to verify: Listen to transcript at 51:23; verify by checking Speaker_0's newsletter or course content for advice on beginner AI usage. For recent events, verify 2026-08-09 through 2026-11-09. Use another period only when the speaker explicitly names it.
Do your own research
- Verify the current capabilities and API limits of Claude regarding Python execution and external data access.
- Review the specific data sources mentioned (EODHD, EDNet, SEC EDGAR) for cost, coverage, and reliability.
- Assess the security implications of storing personal investment decisions and portfolio data in local or cloud-based AI systems.
- Check the accuracy of the claim that 40% of investors never use AI by consulting recent surveys from firms like Morningstar or Pew Research.
- Evaluate the 'vault' concept by reviewing how past investment decisions are structured and stored for AI retrieval.
- Confirm the functionality of Interactive Brokers MCP connections for automated portfolio analysis.
- Research the 'Stairway Framework' steps (prompting, Skills, data, agents, memory) to understand the technical requirements.
- Look for counter-arguments to the 'human-in-the-loop' model, such as cases where automation led to undetected errors.
Full transcriptexpand
- 0:00
- Unidentified speaker (speaker_0): What is important is this is a a picture of how my AI pipeline works. You can see
- 0:05
- Unidentified speaker (UNKNOWN): that there
- 0:06
- Unidentified speaker (speaker_0): are different agents. So hunter searches for new ideas, researcher creates a report, the contrarian will do uh a pushback, and the tracker system will auto track all the holdings I have. But you clearly see that there is always a human involved. So it's never pure AI, it's a human AI system. Because you use an agent, it produces a result,
- 0:26
- Unidentified speaker (speaker_0): it goes to a human. in this case me. I make a decision and then it goes back to the next agent, and so forth, and so forth. The most important is in the middle, it's called the vault. And this is your your knowledge, your expertise, all the decisions you made on investing written down so that the agent can use it to make a better preparation for you, and
- 0:45
- Unidentified speaker (speaker_0): all the agents use the same vault.
- 0:48
- Ian Cassel: This is club conversations powered by Microsoft. MicroCap Club. Our goal is to give our community access to the best minds in investing and business. I'm Ian Castle, founder of MicroCap Club. MicroCap Club is a private community of the world's best stock pickers. Since 2011, our members have profiled 1,500 companies
- 1:08
- Ian Cassel: and over 300 have turned into multi-baggers. That's at least one big winner every month for 14 years. If you enjoy finding great companies early, well before Wall Street pays attention, MicroCap Club is where you belong. I hope to see you in our community. And now a quick disclaimer. This presentation is for informational purposes only and should not be construed as a
- 1:28
- Ian Cassel: recommendation to purchase or sell any security referenced herein. Planet MicroCap Holdings LLC and MicroCap Club LLC LC are not licensed brokers nor registered investment advisors. We, our partners, contractors, members, subscribers, guests, or affiliates may or may not hold positions in one or more of the securities mentioned in this presentation and may trade in such securities at any time. We recommend you consult a licensed investment advisor, broker, or legal counsel before purchasing or
- 1:48
- Ian Cassel: selling any securities referenced in this presentation.
- 1:57
- Unidentified speaker (speaker_2): Capital. Kevin came to markets from nuclear physics and has been investing for 17 years, hunting for overlooked high-growth microcaps around the world through his 100-bagger hunting newsletter. For the past two years, he has built AI into his process from screening entire sectors at once to an agent that tests each of
- 2:17
- Unidentified speaker (speaker_2): his thesis against new filings. And he now teaches other investors to do the same through his newsletter Invest with AI. Kevin walks us through how that system is built. and where an individual investor should start. Let's get into it.
- 2:31
- Unidentified speaker (UNKNOWN): Hi,
- 2:31
- Unidentified speaker (speaker_3): Kevin. Thank you so much for joining us today. I'm glad to have you here. And I know you have prepared some really cool slides for us for the topic of today, which is uh how to use AI in our investment process. And at the end, I'm also happy to ask you some follow-up questions that I have prepared. So yeah, with that
- 2:51
- Unidentified speaker (speaker_3): said, uh the floor is yours to take control of the show.
- 2:57
- Unidentified speaker (speaker_0): Well, uh thanks a lot, David, for having me. Uh yes, I prepared uh some slides. We will go through them and then we can discuss and I will try to answer all your questions. So
- 3:09
- Unidentified speaker (UNKNOWN): let me
- 3:09
- Unidentified speaker (speaker_0): maybe start with uh a short introduction for if you don't know me, so you can find me on X. My name is Kevin. I
- 3:17
- Unidentified speaker (UNKNOWN): I
- 3:17
- Unidentified speaker (speaker_0): run a company called Atomical Capital. Um my background is not in investing. I've been investing for 17 years now, but my background is in uh in nuclear physics. I will come back to that in a moment. Um and as you can see on my profile it says uh hunting hidden atoms. So I also hunt in the microcap
- 3:37
- Unidentified speaker (speaker_0): space, trying to find uh unfollowed companies with with a lot of growth potential in the future. And um yeah, over time I've been uh using AI in my investing process for the last two years.
- 3:51
- Unidentified speaker (UNKNOWN): Uh
- 3:51
- Unidentified speaker (speaker_0): over time I got more and more questions uh from readers on how are you doing this, etc. So I'm also teaching investors now how to use uh how to use cloud. Claude.
- 4:04
- Unidentified speaker (speaker_0): So basically my business is is we I do not have a fund. It's called Atomical Capital based on my background, but um I write uh two newsletters. Uh the first is called Hunterbagger Hunting uh where we we do your typical deep dives and portfolios um hunting for microcaps with high growth
- 4:23
- Unidentified speaker (speaker_0): potential. But then the newer newsletter and and course, because it's a video course I created, is called Invest with AI. And there we talk about yeah how can you use uh in particular Cloud uh to enhance your your investing process.
- 4:43
- Unidentified speaker (speaker_0): Now let me maybe go back to the to the beginning because as I said my background is in physics and and this is uh Professor Richard Feynman. Um when I get the question on um who is your favorite investor, I usually reply with my favorite investor is Richard
- 5:02
- Unidentified speaker (speaker_0): Feynman, although I have no idea if he invested in the past. Um but he won the Nobel Prize for Physics. But what p most people uh don't know is he he was uh an incredible teacher. So his ability was to explain very complex topics in
- 5:22
- Unidentified speaker (speaker_0): a very simple manner. And so my my uh message to you, David, and also to people who are watching or or listening, is if you do not understand everything I thing tell you at the end of this presentation, then that is my fault, it is not yours. I should have explained it. in a simpler manner. Um
- 5:42
- Unidentified speaker (UNKNOWN): now
- 5:42
- Unidentified speaker (speaker_0): one other thing I have to mention about Richard Feynman and why I like him so much for investing in the stock market is he says um always distrust experts and even distrust yourself. Because the basic idea is um he has a a sort of quote uh that he uses or used in the past. Um
- 6:03
- Unidentified speaker (UNKNOWN): which
- 6:03
- Unidentified speaker (speaker_0): says
- 6:05
- Unidentified speaker (speaker_0): you as a human are the easiest to fool.
- 6:09
- Unidentified speaker (speaker_0): That's the first principle, don't fool yourself. And why does it tie to investing? Well,
- 6:15
- Unidentified speaker (UNKNOWN): the
- 6:15
- Unidentified speaker (speaker_0): basic premise in physics is you build up a thesis on something you observe. So, for example, in the case of a company, it's the same thing.
- 6:23
- Unidentified speaker (UNKNOWN): And
- 6:24
- Unidentified speaker (speaker_0): then everything you do afterwards is trying to find evidence to destroy the thesis. So you're always looking for uh the the the short report or the the person who is trying to break your inner uh original thesis.
- 6:39
- Unidentified speaker (UNKNOWN): Okay,
- 6:39
- Unidentified speaker (speaker_0): so that's a bit of the context uh about me. Now um
- 6:45
- Unidentified speaker (speaker_0): the presentation itself is is constructed as a stairway
- 6:50
- Unidentified speaker (UNKNOWN): and
- 6:51
- Unidentified speaker (speaker_0): uh we're just going to go up the five um steps
- 6:55
- Unidentified speaker (UNKNOWN): uh
- 6:55
- Unidentified speaker (speaker_0): pretty quickly. Um it starts with your basic prompting, what most people already know. I'm going to talk to you a little bit about Claude Skills and why it's so useful.
- 7:06
- Unidentified speaker (UNKNOWN): And
- 7:06
- Unidentified speaker (speaker_0): we're going to talk about data and agents, and then the last step is memory, but memory I will not go into depth in in that. The basic idea at the end of the presentation is to show you some actual practical use cases of how you can use AI to amplify or leverage your
- 7:26
- Unidentified speaker (speaker_0): own investing process.
- 7:29
- Unidentified speaker (speaker_0): So let's maybe start at the beginning. So
- 7:32
- Unidentified speaker (UNKNOWN): in
- 7:32
- Unidentified speaker (speaker_0): 2022 we had ChatGPT, and I wanted to be um once these tools came out
- 7:40
- Unidentified speaker (UNKNOWN): I
- 7:41
- Unidentified speaker (speaker_0): was one of the the I wanted to be at the forefront so I started using them immediately and after the initial magical moment uh because it was a bit of a magical moment and I think the the data also supports that because um ChatGPT had 100 million users in only two months
- 7:58
- Unidentified speaker (UNKNOWN): uh
- 7:58
- Unidentified speaker (speaker_0): compared to like Facebook uh did it did it in fifty-four fifty four months But after that initial magic, um for equity analysis, the conclusion was it's a bit beautifully beautifully written nonsense. Uh because you get a feedback from the tool which
- 8:18
- Unidentified speaker (speaker_0): is super confident, which looks which looks amazing and that it sounds amazing, but there is errors in them.
- 8:26
- Unidentified speaker (UNKNOWN): And
- 8:26
- Unidentified speaker (speaker_0): that's not what you want. Now, this is 2022, so it's almost four years ago. Um at the time for myself I had no clue how to use these tools. That's that's first the first thing. And the second thing is the models themselves at that point in time were just not strong enough. So um
- 8:45
- Unidentified speaker (speaker_0): now
- 8:48
- Unidentified speaker (speaker_0): what's very important uh when we're talking about using an LLM
- 8:54
- Unidentified speaker (speaker_0): is uh and this is probably one of the most important slides in the in the in the the short slide deck I'm going to show
- 9:00
- Unidentified speaker (UNKNOWN): you.
- 9:00
- Unidentified speaker (speaker_0): So you see uh on the left is a car and the LLM is the engine. And what's important in the um
- 9:06
- Unidentified speaker (UNKNOWN): in
- 9:07
- Unidentified speaker (speaker_0): this analogy is you are at the steering wheel. And that means when you're using these tools,
- 9:14
- Unidentified speaker (UNKNOWN): you
- 9:15
- Unidentified speaker (speaker_0): always have to have the mindset or the mental model of you are the senior and the LLM is the junior. You are steering and the LLM needs to follow, and you are the judge and the LLM prepares. So you can never take uh a result at point blank range. Um when you get a result from an LLM, uh you are the the
- 9:34
- Unidentified speaker (speaker_0): final decision maker.
- 9:37
- Unidentified speaker (UNKNOWN): Um
- 9:37
- Unidentified speaker (speaker_0): that's very important. Uh because it also means that you are going to steer the chatbot in a certain direction based on what you ask.
- 9:47
- Unidentified speaker (speaker_0): Now, I often get the question because it's moving so fast, all of these things, are we not too late? And so you have this chart where you have um
- 10:00
- Unidentified speaker (speaker_0): the the square brackets that are uh black, these are the free users, for example, for Chat GPT. Uh about one billion users. So you can see 2.5 million users for one um uh square and there's 40 in total. Now
- 10:19
- Unidentified speaker (speaker_0): if you have if you take those that have a paid subscription, that's the blue ones, and then if you take the power users. that's the red ones. And
- 10:30
- Unidentified speaker (speaker_0): the the goal of this slide is to show you that we're still very early. So it's never too late to learn all these things.
- 10:50
- Unidentified speaker (speaker_0): Now, I have done uh these kind of presentations in the past. Um I have about 2,000 people who followed them, and then when I asked the these my previous audience, uh, do you use AI for investing? You can see that 40% never use it, and 37% said occasionally.
- 11:08
- Unidentified speaker (UNKNOWN): And
- 11:09
- Unidentified speaker (speaker_0): then at the right, you can see what what the use cases are. So the most most of the people you use it as like a sort of a replacement for Google search. They say it's for screening stocks, but if I dug deeper, it's more like finding new ideas
- 11:26
- Unidentified speaker (speaker_0): or making summaries or tracking a portfolio that's other use cases.
- 11:30
- Unidentified speaker (UNKNOWN): Now
- 11:30
- Unidentified speaker (speaker_0): I do have to emphasize uh these are a lot of individual investors responded. These are not like uh professional investors or full-time investors. So I do think um the you the um the result would be different if you go into the professional professional sphere.
- 11:48
- Unidentified speaker (speaker_0): Um
- 11:50
- Unidentified speaker (speaker_0): this slide is just to show you that the basic framework is always the same and it doesn't matter what the scale is. What I mean for this is about two weeks ago I had um a meeting with um uh the COO of a big asset management firm and we were discussing about how they are implementing AI into their their business. So they
- 12:10
- Unidentified speaker (speaker_0): have a business with about 2,000 people. Um of course my businesses are much smaller. I am a solo operator uh of newsletters. Um and so he said, Well, look, um it takes about four years. Uh four years ago we started with like um cleaning up our data. And they took two years to make sure that the
- 12:30
- Unidentified speaker (speaker_0): data was of high quality. And then they built an application layer to use like um uh automations without even AI. We're not talking about AI yet, on this data.
- 12:41
- Unidentified speaker (UNKNOWN): And
- 12:42
- Unidentified speaker (speaker_0): they have now started like a year ago implementing AI. Um
- 12:46
- Unidentified speaker (UNKNOWN): of
- 12:47
- Unidentified speaker (speaker_0): course this takes a lot of time. Um the
- 12:52
- Unidentified speaker (speaker_0): The part below where it says small fund or solo investor, it's the exact same process. You have to make sure you have the data, you put an application layer on it where you start reading it, and only then do you apply AI. Of course, you can do this in a matter of weeks or months compared to a big form firm who needs um all kind of enterprise
- 13:12
- Unidentified speaker (speaker_0): security, uh and it takes just a lesson. list a lot of a lot of more time and then the the the line you see increasing it's the moment you start compounding your expertise because one of the use cases and uh what's what the person told me from the asset management form is uh compounding knowledge
- 13:31
- Unidentified speaker (UNKNOWN): because
- 13:32
- Unidentified speaker (speaker_0): an asset management firm uh they have clients so relationships but on the other side they have their uh proprietary data and investing expertise and they want to compound that expertise
- 13:44
- Unidentified speaker (UNKNOWN): and
- 13:45
- Unidentified speaker (speaker_0): it's the same thing for uh for a solo investor or uh or a small fund
- 13:50
- Unidentified speaker (speaker_0): often you compound your knowledge faster than your portfolio sadly but uh that's the that's the truth.
- 13:57
- Unidentified speaker (speaker_0): Now
- 14:00
- Unidentified speaker (speaker_0): let's go back to basics. Um you can see here um at the top right you can see your typical claw chat and when you're talking to an AI tool,
- 14:11
- Unidentified speaker (UNKNOWN): you
- 14:12
- Unidentified speaker (speaker_0): have three components. You have the data,
- 14:14
- Unidentified speaker (UNKNOWN): you
- 14:15
- Unidentified speaker (speaker_0): have the prompt or the question you ask, and you have the model you choose or the brain. So three building blocks the data, the brain, and the prompt. Um in the first situation when you have great data
- 14:29
- Unidentified speaker (UNKNOWN): and
- 14:29
- Unidentified speaker (speaker_0): you have a very smart brain,
- 14:31
- Unidentified speaker (UNKNOWN): but
- 14:31
- Unidentified speaker (speaker_0): you give it a vague prompt, then you still get a decent result. Now why is that? It's because the intelligence, the models have become so good that they can compensate for the vagueness of your prompting. Okay.
- 14:46
- Unidentified speaker (UNKNOWN): Now
- 14:46
- Unidentified speaker (speaker_0): in the second situation, we have g great data.
- 14:49
- Unidentified speaker (UNKNOWN): You
- 14:49
- Unidentified speaker (speaker_0): use a dumber brain.
- 14:51
- Unidentified speaker (speaker_0): but you use a fantastic prompt. Well, in reality, it could result in an even weaker result. Why? Because the brain has become so important that if you choose a dumb brain, it will not reliably execute your advanced prompt you gave
- 15:09
- Unidentified speaker (UNKNOWN): it.
- 15:10
- Unidentified speaker (speaker_0): So in some situations it's even weaker than the first uh line. And then at the end, um, you have bad data with a smart brain and a good prompt. And I mentioned on the slide here, yikes. Why? Because it's gonna it's going to give you a super confident reply, but
- 15:30
- Unidentified speaker (speaker_0): because of the bad data, it will be incorrect.
- 15:33
- Unidentified speaker (UNKNOWN): And
- 15:33
- Unidentified speaker (speaker_0): that's the biggest danger you can have. And of course, what we want is you want all three building blocks to be. green.
- 15:42
- Unidentified speaker (speaker_0): So that was the introduction. Now let's go into skills.
- 15:48
- Unidentified speaker (speaker_0): Um
- 15:48
- Unidentified speaker (UNKNOWN): first
- 15:49
- Unidentified speaker (speaker_0): things first. So um what what are the possibilities of a large language model? A large language model basically is statistical.
- 15:57
- Unidentified speaker (UNKNOWN): So
- 15:58
- Unidentified speaker (speaker_0): if you give something a prompt
- 16:00
- Unidentified speaker (UNKNOWN): and
- 16:00
- Unidentified speaker (speaker_0): you do it ten 10 times, it will give you the output will be a bit different every time. That's just the nature of how it works. It's a mathematical model. Now, on the picture here, you can see my son, he's 10 years old, and he started school uh again this year. Um, now he's super uh strong in math, so he doesn't have to
- 16:20
- Unidentified speaker (speaker_0): do anything special, he just he's strong at it. But for the first time this year, he needed to learn. French. French is a bit more difficult. Now he didn't allow me to put the
- 16:32
- Unidentified speaker (UNKNOWN): the
- 16:32
- Unidentified speaker (speaker_0): square in red, but he said, okay, orange is okay, so it's it's okay.
- 16:37
- Unidentified speaker (UNKNOWN): You
- 16:38
- Unidentified speaker (speaker_0): have to understand that a large language model um it is very strong at language, but it is not good at all at math. So when a large language model starts doing calculations, maybe for a evaluation, you need to pay attention how it is performing the calculation.
- 16:57
- Unidentified speaker (speaker_0): Now, what is a skill? I show you, we talked about the prompts, so the questions you ask in the chatbot. A skill is basically your experience, your process. It's is a small piece of software. Um You are going to teach clause. Clauds how to do, for example, a lifecycle
- 17:17
- Unidentified speaker (speaker_0): analysis evaluation. Um you can make uh for example a competitive advantage analysis or an industry peer analysis. These are all all skills that you can build so that so that you can you build it once and then you can just apply it every time on any company. Um the basic idea is if you're
- 17:37
- Unidentified speaker (speaker_0): using these tools and you find yourself using the same prompts all the time, then it will probably it probably needs to become a skill.
- 17:57
- Unidentified speaker (speaker_0): Well, you can use or build a lifecycle skill, and then it can produce something like this, where it shows you where the company sits in its life cycle. Now, what is important here, because we talked about it before, um, you can see inside this skill it's not just a prompt, it includes Python scripts. And
- 18:17
- Unidentified speaker (speaker_0): that is for me the the biggest unlock when using Claude is you can use Python. You don't even have to know Python because Claude knows Python.
- 18:26
- Unidentified speaker (UNKNOWN): And
- 18:27
- Unidentified speaker (speaker_0): it's important because, as we said, Claude itself or an LLM is statistical, and Python allows you to
- 18:37
- Unidentified speaker (speaker_0): bring more determinism into the system.
- 18:40
- Unidentified speaker (speaker_0): Often I relate, I don't know if you know the movie Determinator with Arnold Schwarzenegger, Um, but often I I do the analogy that
- 18:50
- Unidentified speaker (UNKNOWN): the
- 18:51
- Unidentified speaker (speaker_0): Terminator itself, Arnold Schwarker Schwarzenegger, he has a metal endoskeleton, and that's always fixed. But in the second movie, it's a T1000 with Roger Patrick, and he has liquid metal,
- 19:03
- Unidentified speaker (UNKNOWN): and
- 19:03
- Unidentified speaker (speaker_0): that's the LLM. And so the best combination is both. That's what we're doing here with these kind of skills.
- 19:12
- Unidentified speaker (speaker_0): Um, and so typically uh I've constructed a video course with a skill library and you can download these kind of skills from it and it's always a combination of Python and uh
- 19:23
- Unidentified speaker (UNKNOWN): the
- 19:23
- Unidentified speaker (speaker_0): LLM itself with the with the markdown file.
- 19:28
- Unidentified speaker (speaker_0): Okay, I went to skills pretty fast. Let's take a look at the data. Now what's important when using this for stock analysis is of course your context. Now I'm just going to show you one slide on how, for example, when I use Cloud Cowork, how it is connected to my life and my data.
- 19:47
- Unidentified speaker (UNKNOWN): And
- 19:47
- Unidentified speaker (speaker_0): it looks like this. I pull um, for example, data from um a provider called EODHD for the financial data itself. And then I use another provider called Financial Filings for the filings. Because you always need to make sure that you have the filings themselves and
- 20:07
- Unidentified speaker (speaker_0): the financial data itself, like in an Excel file or something, or separate. Because if you give the LLM a PDF and you just you just think okay it will extract the data correctly, there can be errors in it. Because if it has if it has to extract text that's easy.
- 20:26
- Unidentified speaker (UNKNOWN): If it
- 20:26
- Unidentified speaker (speaker_0): has to extract a table in the right structure, there can be errors inside. So I always prefer to have the financial data separate from the filings themselves.
- 20:37
- Unidentified speaker (UNKNOWN): And
- 20:37
- Unidentified speaker (speaker_0): then you can see I have my Gmail and my Notion linked. And you can also see my smartphone because everything in my process I can uh use it from my smartphone. So for example, if I'm at a conference and I see an interesting stock, then I will just talk into my smartphone and ask Claude to do uh a research analysis on uh
- 20:57
- Unidentified speaker (speaker_0): a ticker, and then when I come home I have a research report ready based on my custom process. It's not a general research uh function, but based on my process.
- 21:09
- Unidentified speaker (speaker_0): This is, for example, um a specific skill that Anthropic created with some financial institutions like JP Morgan. It's a skill that is free. You can download it from the internet. And it produces an uh initiating coverage report. It's uh used on the sell site. Uh it's like uh a complete
- 21:29
- Unidentified speaker (speaker_0): research report on a company, about 50 pages. And it's interesting because um in a very easy way you can have um uh a lot of data ready before starting your own analysis. Now, the only problem with this is you run this, it takes it 30 minutes, about
- 21:49
- Unidentified speaker (speaker_0): 30 minutes to create your report. You have to you don't have to do anything, it will do everything for you. But then of course the big question afterwards is um what's what about the quality? I mean, will there be hallucinations inside or not? Um and what we did was we added uh some sort of uh an
- 22:08
- Unidentified speaker (speaker_0): independent quality control so that you get um reliability scoring, and then you just have to look at what is yellow. Um it will screen the entire report, and then you just check the yellow uh warning signs to see if there's any problem, if for you it's acceptable or not to to to keep the report as is.
- 22:30
- Unidentified speaker (speaker_0): Everything I discussed up until now, you can just do on your computer with a desktop computer. You don't need anything special.
- 22:38
- Unidentified speaker (speaker_0): Just want to show that for my setup I use a Mac Mini
- 22:43
- Unidentified speaker (speaker_0): because it's my personal server, but you don't really need this. I just mention
- 22:48
- Unidentified speaker (UNKNOWN): it
- 22:48
- Unidentified speaker (speaker_0): because this is one of the use cases.
- 22:51
- Unidentified speaker (speaker_0): Okay. Um I went fast through all the different steps because I want to talk about agents. Uh that's we are now in the agentic era in uh in AI um in the AI era. And I think it's important to show you what is possible uh when you have an agent at your disposal. So, first of
- 23:11
- Unidentified speaker (speaker_0): all, uh the agents of the market, so Cloud Cowork is an aided agent agent for knowledge work, so basically it's perfect for equity analysis. Um, if you want to be able to go a bit deeper, uh you can of course uh use Cloud Code. Um
- 23:28
- Unidentified speaker (speaker_0): the basic distinction is Cloud Cowork is for doing stuff and Cloud Code is for building stuff. So if you want to build your own investing tools, you can do that with Cloud Code. Of course, there are alternatives like Perplexity Computer and ChatGPT Codex and a couple of others.
- 23:45
- Unidentified speaker (UNKNOWN): The
- 23:45
- Unidentified speaker (speaker_0): basic idea for an agent is it runs in the loop and it will perform a task for you and it will try different things to achieve that task up until maybe there's a timeout or until it has achieved the task.
- 24:02
- Unidentified speaker (speaker_0): What is important is this is a a picture of how my pipeline, my my AI pipeline works. What's important here is that you can see that there are different agents. So hunter searches for new ideas, researcher creates a report,
- 24:18
- Unidentified speaker (UNKNOWN): the
- 24:19
- Unidentified speaker (speaker_0): contrarian will do a pushback,
- 24:21
- Unidentified speaker (UNKNOWN): and
- 24:21
- Unidentified speaker (speaker_0): the tracker system will auto-track all the the the holdings I have. but you clearly see that there is always a human involved. So it's never pure AI, it's a human AI system.
- 24:35
- Unidentified speaker (speaker_0): Because you use an agent, it produces a result, it goes to a human, in this case me, I make a decision, and then it goes back to the next agent, and so forth, and and so so forth. forth. Um The most important is in the middle, it's called the vault. and this is your your knowledge, your expertise,
- 24:54
- Unidentified speaker (speaker_0): all the decisions uh you made on investing written down so that the agent can use it uh to make a better preparation for you and all the agents use the same vault.
- 25:09
- Unidentified speaker (speaker_0): Um
- 25:10
- Unidentified speaker (UNKNOWN): now
- 25:11
- Unidentified speaker (speaker_0): let's get into some um very specific examples So the first example is finding ideas. So
- 25:20
- Unidentified speaker (speaker_0): you probably know when you're hunting at microcaps,
- 25:23
- Unidentified speaker (UNKNOWN): you
- 25:24
- Unidentified speaker (speaker_0): have to go through a lot of stocks. Usually a lot of investors they go A to Z through an exchange to find hidden treasures inside. Um
- 25:35
- Unidentified speaker (speaker_0): With an agent, you can do sector analysis. Uh so you can like analyze 100 to 200 companies in one go. Um
- 25:45
- Unidentified speaker (UNKNOWN): now
- 25:46
- Unidentified speaker (speaker_0): why is that you can use financial data, that's not new because in the past, when you use a stock screener, you can just say I want everything at twenty percent revenue and I don't know, maybe ten 10% percent uh RIC and give me all the companies. What's special here is that you're going to go through all the filings.
- 26:04
- Unidentified speaker (UNKNOWN): And
- 26:05
- Unidentified speaker (speaker_0): so, for example, uh, the example I've chosen here is I did an analysis of um all the banks in Europe, or at least I took 180 of
- 26:15
- Unidentified speaker (UNKNOWN): them.
- 26:15
- Unidentified speaker (speaker_0): Um the first step was to create a framework to explain to the AI how it needs to analyze a bank and how it needs to rank or give a scoring to a bank.
- 26:27
- Unidentified speaker (speaker_0): Then you collect all the data and the filings, you do a quality control of the data itself to make sure that there is no issues, and then you start the analysis. After the first iteration, it will give you a table like this. And then the next couple of hours is just iterating, asking
- 26:47
- Unidentified speaker (speaker_0): additional questions so that you can ensure that the quality of the end result is okay.
- 26:54
- Unidentified speaker (UNKNOWN): And
- 26:55
- Unidentified speaker (speaker_0): then you get a ranking with 180 banks, and uh and you can then go deeper into one bank and and and try to make a decision if you want to buy it or not. This is something that would take an enormous amount of time without AI. And um I do get the question sometimes what yeah, but this is like a surface.
- 27:15
- Unidentified speaker (speaker_0): analysis. Um
- 27:17
- Unidentified speaker (UNKNOWN): of
- 27:17
- Unidentified speaker (speaker_0): course it's not a deep dive on each bank, but you do take chunks of the text inside the annual reports of the banks and it is included in the analysis. So it's like replicating what a human would do if it would if a human would look at certain specific um chapters or paragraphs inside a of filing,
- 27:37
- Unidentified speaker (speaker_0): it will take it into account.
- 27:41
- Unidentified speaker (speaker_0): Um
- 27:41
- Unidentified speaker (UNKNOWN): the
- 27:42
- Unidentified speaker (speaker_0): same thing when we had the SaaS pocalypse. Um it was in March of this year, I think. So I was thinking, okay, let me just take one hundred SaaS companies in the US. We developed a framework for uh to measure AI resilience uh to see uh can the company be disrupted because of uh the the the new
- 28:02
- Unidentified speaker (speaker_0): AI tool. tools or not
- 28:04
- Unidentified speaker (UNKNOWN): and
- 28:04
- Unidentified speaker (speaker_0): then again you can rank one hundred companies and then dig deeper into for example the the top ones typically what we did was we didn't necessarily want to look at um the best so those who are most AI resilient why because usually the those are already priced in by the market but you want something that is
- 28:24
- Unidentified speaker (speaker_0): just resilient so not the top but not bad, at an attract attractive price. And that's one way to do it. So these are two examples where you're going to do a triage of an entire sector. Um these are just some examples. So when we did it in March,
- 28:41
- Unidentified speaker (UNKNOWN): so
- 28:41
- Unidentified speaker (speaker_0): we had one agent who analyzed 100 SAS companies, uh, and these are just three examples. You have uh Rubrik, Adobe, and Wix,
- 28:49
- Unidentified speaker (UNKNOWN): and
- 28:49
- Unidentified speaker (speaker_0): then you can see the current price performance. Now, of course, it's only six months, so I'm not going to say that there is a true um correlation between the AI scoring in the and the performance itself. The AI scoring is a scoring that is based on multiple years into the future, uh, based on the resilience of the company at
- 29:09
- Unidentified speaker (speaker_0): this time. Um,
- 29:11
- Unidentified speaker (UNKNOWN): Now,
- 29:12
- Unidentified speaker (speaker_0): of course, I didn't buy any of them because these are not microcaps. but it's just to give you an example of uh what you can
- 29:18
- Unidentified speaker (UNKNOWN): do. Uh
- 29:19
- Unidentified speaker (speaker_0): and the second, I think the most basic um or the most obvious use case is thesis tracking. So this is for example an example of
- 29:31
- Unidentified speaker (UNKNOWN): a
- 29:31
- Unidentified speaker (speaker_0): portfolio, and uh normally it's a video because you can't sh see it here. But
- 29:36
- Unidentified speaker (UNKNOWN): what
- 29:36
- Unidentified speaker (speaker_0): you do is when you buy a company. you write down a thesis, and in the thesis you're going to explain why you're buying
- 29:43
- Unidentified speaker (UNKNOWN): it.
- 29:44
- Unidentified speaker (speaker_0): And then what you can do is um, for example, once a day or once every week, you pull in all the new information available on the company and you let it test your thesis. And then once it does the analysis, it will send me an email, and in the email it will be a report. for each position
- 30:04
- Unidentified speaker (speaker_0): position where it says the thesis is still intact, this thesis is under pressure, or it's broken. Now, just to emphasize that this sometimes works too well, is I had a position last year in uh Sunavave, uh Sunwave Sun Away medical, Fnatical. and um
- 30:24
- Unidentified speaker (speaker_0): I think it peaked around September October in 2025. And then in November, based on the new information we had, I think it was from an earnings release, suddenly my system told me the thesis was broken.
- 30:40
- Unidentified speaker (speaker_0): But uh of course you you always do the own judgment. So you take the analysis and then you use the own judgment and based on the information I have I had at the time, I said, Yeah, okay, uh, there are some signs that there might be some cracks in the thesis, but to me, it's not broken, it's under pressure. And so
- 31:00
- Unidentified speaker (speaker_0): I kept it, and the stock kept on dropping. And in the consequent months afterwards, bad news continued to stream in, the stock continued to drop,
- 31:14
- Unidentified speaker (UNKNOWN): my
- 31:14
- Unidentified speaker (speaker_0): TC. tracker was constantly telling me that the TC was broken. In the end, I had to sell because uh my fault was I under investigate underestimated the the impact of uh the change in reimbursement uh regulation on the entire sector because it was not only the company itself.
- 31:32
- Unidentified speaker (speaker_0): Um and I missed that. And that's why
- 31:37
- Unidentified speaker (UNKNOWN): if
- 31:37
- Unidentified speaker (speaker_0): I go back to Feynman, who says Don't fool yourself, if I look back at my seventeen years of investing, um probably the one of the mistakes I made was to hold on too long to situations where I was justifying for myself to keep on uh to not sell the
- 31:56
- Unidentified speaker (speaker_0): company.
- 31:57
- Unidentified speaker (UNKNOWN): And
- 31:57
- Unidentified speaker (speaker_0): for me this is a good system because it's like my bullshit meter. It it throws things back at me. and it says, Yeah, I think you're wrong, and this is why.
- 32:08
- Unidentified speaker (speaker_0): Um
- 32:09
- Unidentified speaker (speaker_0): this is just to show you how it works. But
- 32:13
- Unidentified speaker (UNKNOWN): uh
- 32:14
- Unidentified speaker (speaker_0): it's basically collection of information, uh reading the information, making a judgment. The most important factor in if you want to create a siltup system for yourself is the quality of your thesis. Because uh um if your thesis is vague, the LLM cannot judge.
- 32:31
- Unidentified speaker (UNKNOWN): Uh
- 32:31
- Unidentified speaker (speaker_0): so you have to you have to include for example what I call exit rules. So when you write your thesis to buy a company, you also need to write down in which conditions you think you want to sell it.
- 32:45
- Unidentified speaker (speaker_0): And then it can start doing real judgment.
- 32:51
- Unidentified speaker (speaker_0): Okay, um we're almost at the end. Um so when it comes to memory and context, um
- 32:58
- Unidentified speaker (speaker_0): you also you all know maybe the story of Aladin. Um so normally what you do is you rub the lamp, and then when the genie comes out, you make three wishes. Um but in the case of an LLM, if the genie is the the all-knowing all knowing intelligence you're using, you need
- 33:14
- Unidentified speaker (UNKNOWN): to
- 33:14
- Unidentified speaker (speaker_0): keep it in the lamp. The basic idea is uh an LLM is pre trained on on the the what is available on the internet.
- 33:23
- Unidentified speaker (UNKNOWN): It
- 33:23
- Unidentified speaker (speaker_0): has a cutoff date in its training. If you use it to ask general questions, it will also give you it's better now and in the past, but it also will give you a more have generalized response. If you give it the right context and you focus it, like giving it a specific role, giving it data, it will give you
- 33:43
- Unidentified speaker (speaker_0): a much better answer in your information. And of course,
- 33:47
- Unidentified speaker (UNKNOWN): as
- 33:47
- Unidentified speaker (speaker_0): I said at a certain point in time, the holy grail is to give it all your context. So, for example, if you are someone who has been investing for 20 years and you have been very diligent in writing down all your decisions, why you made it by a this? buy decision, why you made a sell decision, why you trimmed, why you added, etc., and you all always
- 34:07
- Unidentified speaker (speaker_0): wrote down why, then this would be an excellent database you need to give to an LLM. Because it will help you with future decisions. And that's basically what, for example, an asset management firm wants to do. They want to track the decisions the analysts are taking and then use the LLM as a feedback loop.
- 34:27
- Unidentified speaker (speaker_0): to help the analyst, analyst, or portfolio manager to make better decisions.
- 34:33
- Unidentified speaker (UNKNOWN): If
- 34:33
- Unidentified speaker (speaker_0): you set this up with the tools you can have at home, like with Cloud Corework on a desktop, it means you're going to give it a brain, and the brain is just a bunch of folders on your computer with files in it,
- 34:45
- Unidentified speaker (UNKNOWN): and
- 34:46
- Unidentified speaker (speaker_0): it will read and write to those folders. That's the most the simplest system you
- 34:50
- Unidentified speaker (UNKNOWN): can do.
- 34:51
- Unidentified speaker (speaker_0): have to build a brain.
- 34:55
- Unidentified speaker (speaker_0): Um
- 34:56
- Unidentified speaker (UNKNOWN): yeah,
- 34:57
- Unidentified speaker (speaker_0): and this is one of the examples. So I like I said, learn from your own decisions, uh, the miracle of compound interest apply to your own investing journey. I'm just going to give one example and I experienced it about three or four weeks ago.
- 35:09
- Unidentified speaker (UNKNOWN): Uh
- 35:09
- Unidentified speaker (speaker_0): I had to do an earnings update on one of my companies, and um
- 35:15
- Unidentified speaker (UNKNOWN): I
- 35:15
- Unidentified speaker (speaker_0): have been so, for example, everything I built, I keep building on it every week, so it gets stronger every week. And like when I was doing the earnings updates,
- 35:26
- Unidentified speaker (speaker_0): I use an agent with Claude to help me prep all the information. It reads my past statements Last on it. And suddenly in the analysis, it started quoting what I wrote. in the past. And it started telling me that I made a mistake.
- 35:43
- Unidentified speaker (UNKNOWN): And
- 35:44
- Unidentified speaker (speaker_0): it flagged a mistake I made about one year earlier in my analysis.
- 35:49
- Unidentified speaker (speaker_0): And it copies and pasted paragraphs that I had written to counter argument my own analysis.
- 35:58
- Unidentified speaker (UNKNOWN): And
- 35:59
- Unidentified speaker (speaker_0): that for me that was a new like magical moment because that was the first time. I experienced that. And it shows that the models are getting stronger and that if you have a good pipeline you can get these kind of results. And the basic idea is always that you want
- 36:14
- Unidentified speaker (UNKNOWN): if
- 36:14
- Unidentified speaker (speaker_0): you want to build a junior analyst for yourself, you don't want to build an analyst that always agrees with you. You want somebody that pushes back against your own thinking.
- 36:25
- Unidentified speaker (UNKNOWN): I
- 36:25
- Unidentified speaker (speaker_0): just included this because this is the basic principle of how Andre Carpathy, who worked at uh OpenAI uh and now at Anthropic, how he builds um a second brain or or how he structures the knowledge you have. I just included it in the slideshow. Um it could be interesting if you want to start building this.
- 36:45
- Unidentified speaker (speaker_0): So in this presentation we climbed the stairs from prompt to uh how to include the memory. I just want to end this short presentation with uh to go back to Richard Fyman. You must not fool yourself and you are the easiest person to fool. Um so yeah, that's the way you need to use all this
- 37:04
- Unidentified speaker (speaker_0): AI tooling is to um try to push back on your own thesis and make sure that everything you build is built to help you, to prepare for you. but the judgment always stays with the human.
- 37:33
- Ian Cassel: Toronto. It's where the microcap community connects and great ideas are found. Don't just listen to the conversation, be a part of it. Secure your seat today. Head over to planetmicrocap.com right now to register. See you there.
- 37:48
- Unidentified speaker (speaker_3): First of all, let me say that this was extremely, extremely helpful. It just made me feel like I have a lot to improve in terms of implementing AI in my process. I already really think that I'm pretty active and have initiative to uh sort of try new things. And I have some projects on Cloud that help me
- 38:08
- Unidentified speaker (speaker_3): and have some context on my investing process and decisions. But I think you took it to the next level. And so I feel like this will be extremely helpful for everyone and also for me as well. So I appreciate and want to to thank you for sharing everything that you have done so. far when it comes
- 38:27
- Unidentified speaker (speaker_3): to AI. And and I can understand why all the people that have listened to your presentation have uh really enjoyed it. And I'm not saying this because you're uh you're a guest on the podcast, but I I truly think that you you've done something special here. So I really appreciate that. With with that said, I I do have a couple of
- 38:47
- Unidentified speaker (speaker_3): follow-ups for uh people that want to take this into the practical level. So one thing that I really uh want to implement from what you just said is the the thesis tracker. Because I feel like if you're a solo investor and you don't have a proper structure, it's really easy to sort
- 39:07
- Unidentified speaker (speaker_3): of miss something that changed, or having too many positions to track and you can't really focus on every single one of them at the same time. with the same sort of quality of information and effort. So
- 39:20
- Unidentified speaker (UNKNOWN): uh
- 39:21
- Unidentified speaker (speaker_3): guide us through how how do you source the information to feed the thesis tracker? Because you mentioned you do upload it or update the information weekly or whatever you you choose to do that.
- 39:36
- Unidentified speaker (speaker_0): Well, you have two ways. Um you can rely on free information online because because you are investing in public companies, you can
- 39:45
- Unidentified speaker (UNKNOWN): in
- 39:46
- Unidentified speaker (speaker_0): a certain way retrieve all the information you want through the internet.
- 39:50
- Unidentified speaker (speaker_0): Now, the beauty in the United States is the Edgar system uh from the SEC where you can get the filings. Cloud Clot can like easy easily access it.
- 40:00
- Unidentified speaker (UNKNOWN): It
- 40:00
- Unidentified speaker (speaker_0): has an API at everything, so when it comes to US companies, you can like pull in all the filings. And in the TC Stracker I built, of course, this just happens every Monday at seven o'clock in the morning. He will pull in all the filings or all the news or all the new information. If you want to go
- 40:19
- Unidentified speaker (speaker_0): beyond the US, for example, Europe, Australia, Asia, etc., there are other possibilities, for example Japan, their API is called EDNet. And also there you can just pull in all the filings you want and you have all the information. Now I want to emphasize that for
- 40:39
- Unidentified speaker (speaker_0): my use case, I do use a paid data provider.
- 40:43
- Unidentified speaker (speaker_0): Do you really need it? It's Maybe not, but it just makes life easier. Because they have all the data and you're just going to pull it in every week. So that's that's how you can set it up.
- 40:55
- Unidentified speaker (speaker_3): And what sort of data providers do you recommend and how do you link that to your cloud application?
- 41:03
- Unidentified speaker (speaker_0): Well, what's what what's really useful when you use cloud, for example, is you can connect almost any application. So I think the last time I looked, so they have a customized function where you can add connectors. I think there were about 1500 or maybe 2000, I'm not sure, applications that have provided a connector to Cloud.
- 41:23
- Unidentified speaker (speaker_0): And so this is not a recommendation for what I'm using or anything, but I use um a European data provider called EODHD. It was uh shown in the slides. And you just if you have a subscription to that data provider, you just go into Cloud, you add the connector, you log in with the your user ID and your
- 41:43
- Unidentified speaker (speaker_0): password for your data provider, and then you can just start pulling in the data. So it's as easy as that. More and more of these providers are providing an API or an MCP server, as they call it. Other alternatives are or fiscal.ai. You have financial modeling prep, they also have an API and an MCP. So you
- 42:03
- Unidentified speaker (speaker_0): have different options. The the reason why I use EODHD is because I invest in microcaps worldwide. For the moment, for me, they have the the the broadest set of data.
- 42:17
- Unidentified speaker (speaker_0): Now in the in the structure, the it's like a waterfall. So you you you use in the in the way Claude gets the information,
- 42:26
- Unidentified speaker (UNKNOWN): you find
- 42:26
- Unidentified speaker (speaker_0): first try the connector, and then we do a check to see, okay, did I pull in all the data? Because, for example, it is possible that you want let's say an investor presentation.
- 42:38
- Ian Cassel: And
- 42:38
- Unidentified speaker (speaker_0): the investor presentation is available on the website of the company. In that case, you just need to tell Claude, okay, every week at seven o'clock, if you pull in the data, also check the website of the company. So it's not the data provider on itself is not enough if you want to have a complete set of data. Now,
- 42:58
- Unidentified speaker (speaker_0): why do I emphasize this?
- 43:07
- Unidentified speaker (speaker_0): Because just like a human, if you do a prediction on a company, if you're doing an analysis, you want to have if they did if they did a an interview, for example, last week, like the CEO, you want that interview, you want the transcript, and that's very important.
- 43:21
- Unidentified speaker (speaker_3): Hmm. That's very helpful. And I I also heard or saw in my interactive broker's account that there's a there's an MCP connection as well to Cloud. I'm curious to know whether you have experimented with that and whether it has been helpful and how do you leverage that? in your process?
- 43:40
- Unidentified speaker (speaker_0): Yeah, so I did create a video lesson on that. Basically, you're going to connect your live portfolio to Cloud
- 43:49
- Unidentified speaker (speaker_0): and you can start asking questions about your portfolio, about your portfolio position. Now, what's interesting with that is you can go beyond what Intractor Interactive Brokers gives you because imagine that you're saying, okay, uh pull all the recent portfolio data from my Interactor Brokers account, it will pull it
- 44:08
- Unidentified speaker (speaker_0): in into a chat window in Cloud, but then you can start analyzing the data. And so, for example, you can say, Okay, based on my portfolio at the moment, do me a specific analysis on how exposed post I am to rising interest rates, rate, for example. And yeah, like you can ask several questions. about your portfolio
- 44:28
- Unidentified speaker (speaker_0): composition. Now I am not somebody who follows the macro very closely because I am a bottle-up investor and I'm pure stock picker. So I'm going company by company. But it is interesting, for example, to analyze your portfolio to get a an overview of what are the biggest risks of your entire portfolio. So for
- 44:48
- Unidentified speaker (speaker_0): example, I have had a friend who did this who is
- 44:52
- Unidentified speaker (speaker_0): he has a couple of tech exposure but but most of his portfolio is not tech.
- 44:58
- Unidentified speaker (UNKNOWN): And
- 44:58
- Unidentified speaker (speaker_0): he did the analysis and the result was that he was surprised to see that almost sixty percent of his portfolio is exposed to AI Capex spending. And he he he was not aware of that when when he when he picked the stocks individually.
- 45:14
- Unidentified speaker (speaker_3): Mm-hmm.
- 45:14
- Unidentified speaker (speaker_0): So
- 45:15
- Unidentified speaker (UNKNOWN): that's
- 45:16
- Unidentified speaker (speaker_0): why
- 45:16
- Unidentified speaker (UNKNOWN): it
- 45:16
- Unidentified speaker (speaker_0): could be interesting.
- 45:17
- Unidentified speaker (speaker_3): Yeah, I think it's it's helpful to to sort of get a bigger picture or consolidated consolidated view of your portfolio. Because as you mentioned, like of course we are like bottom up bottom up investors and we focus on the micro, but at the end of the day you're you are also sort of building a yeah, portfolio of companies.
- 45:37
- Unidentified speaker (speaker_3): So there there will be some some useful insights in in there so that makes a lot of sense one other one other question I have is that Claude when when I let's say ask Claude to give me like because I also have a skill that's the a standardized skill of adjustments to
- 45:57
- Unidentified speaker (speaker_3): calculate like EV to EBITDA for example. So for example for European companies it uh it's it strips out the leases and stuff. What happens is that when I forgot to give Cloud the stock price, uh it will probably use an outdated stock price and and then uh I have to correct it. So have you found a fix for feeding live stock
- 46:17
- Unidentified speaker (speaker_3): data into Cloud or is it also coming from the same provider you mentioned before?
- 46:22
- Unidentified speaker (UNKNOWN): Yeah,
- 46:23
- Unidentified speaker (speaker_0): so if you have a paid connector, then it should pull the the most recent data. It could be that there is like a delay for like ten minutes or something because it's not like real life price data, but it's good enough uh for for what you need. If you don't have a connector, normally what Claude does is he will do a web search.
- 46:42
- Unidentified speaker (UNKNOWN): But
- 46:42
- Unidentified speaker (speaker_0): then it's possible that it's not the most recent price as you say. But normally if you ask it to use Yahoo Finance.
- 46:51
- Unidentified speaker (speaker_0): it should be able to get you a price that is closest to reality.
- 46:57
- Unidentified speaker (speaker_3): Mm-hmm.
- 46:58
- Unidentified speaker (speaker_0): Yeah.
- 46:59
- Unidentified speaker (speaker_3): Got it. Got it.
- 47:00
- Unidentified speaker (speaker_0): But the best the best way is to connect to data provider. Yeah.
- 47:05
- Unidentified speaker (speaker_3): And what other is connectors are do you feel are are the highest leverage connectors for people to to explore?
- 47:15
- Unidentified speaker (speaker_0): Yeah.
- 47:16
- Unidentified speaker (UNKNOWN): It
- 47:16
- Unidentified speaker (speaker_0): depends a bit on the use case. So data is a no-brainer.
- 47:21
- Unidentified speaker (speaker_0): There are tools like portfolio trackers. So interactive brokers is a connection to a broker. But for example, I I have four brokers.
- 47:31
- Unidentified speaker (UNKNOWN): Over the
- 47:31
- Unidentified speaker (speaker_0): 17 years, I mean I changed brokers because at a certain point I I I went to microcaps and so N ended on in interactive brokers, but I still have some uh positions in old brokers and then you can use a tool like uh
- 47:47
- Unidentified speaker (UNKNOWN): yeah,
- 47:47
- Unidentified speaker (speaker_0): it's called Portfolio Dividend Tracker or for example Share Site. These are like aggregators that pull in the broker data and they prov also provide connectors.
- 47:57
- Unidentified speaker (UNKNOWN): So
- 47:58
- Unidentified speaker (speaker_0): in that way you can have your full portfolio present even if it's spread over different brokers. And then besides of that, the most important thing for me is to connect like your Google Drive, your Gmail, all your email accounts. Because with all the automations I made, I I let the system email me. Or for
- 48:17
- Unidentified speaker (speaker_0): example, if I have to do
- 48:20
- Unidentified speaker (UNKNOWN): I
- 48:20
- Unidentified speaker (speaker_0): don't know, uh an earnings update analysis, I let it create a Google Doc, for example.
- 48:25
- Unidentified speaker (speaker_0): So those are all different tools, but it it really depends on your workflow. You need to connect the apps you already use so that you can find more leverage through through the use of clouds.
- 48:37
- Unidentified speaker (speaker_3): That's very helpful. And one one thing I really liked about your presentation is the driver engine analogy. Because
- 48:47
- Unidentified speaker (UNKNOWN): one
- 48:47
- Unidentified speaker (speaker_3): of the mistakes I've done myself in fairly and recently actually is that sometimes I delegated the or I try to delegate part of the thinking to to Claude or most of the the task to Claude. And first of all, if you don't have the the right
- 49:07
- Unidentified speaker (speaker_3): system of skills, agents, and and prompting and all of that, the output will be very poor, as you showed in the other matrix where where you had the three scenarios of like poor brain and poor data, and then the poor prompt, or like a combination of those. So I I yeah, the point is to say that uh I
- 49:27
- Unidentified speaker (speaker_3): think that's very helpful and that you should definitely be the driver. But asking about more practical advice on how to how do I say this? How can you make sure you are the driver and not like gradually sort of starting to delegate thinking into cloud?
- 49:47
- Unidentified speaker (speaker_3): Because I I see the main risk of your of your system, and it's a wonderful system. But the main risk I see for me, for example, is that I would start to trust it so much uh that maybe I wouldn't know how to distinguish what's my thinking and what's cloud. Claude's thinking. I I don't even know if you can say Claude's Claude things or not,
- 50:07
- Unidentified speaker (speaker_3): but I think that's debatable. But my qu my question in simple terms is essentially how do you stay the driver in the in the car and not lose control, so to speak.
- 50:17
- Unidentified speaker (UNKNOWN): Yeah
- 50:17
- Unidentified speaker (speaker_0): yeah yeah yeah that's a great question. Yeah so maybe to go back to to the presentation so an LLM is statistical so basically what it is is
- 50:27
- Unidentified speaker (speaker_0): an advanced word guessing machine.
- 50:31
- Unidentified speaker (speaker_0): Now, of course, we haven't figured out exactly how the brain of a human thinks. So maybe when all the studies are over, it maybe you are very similar. I don't
- 50:43
- Unidentified speaker (UNKNOWN): know.
- 50:44
- Unidentified speaker (speaker_0): But for example, I I did a presentation in a university uh last week. There were two 250 students. And I asked them how many of them were using AI tools. And a lot of them are using them more or less, but very few of them had a paid
- 51:03
- Unidentified speaker (speaker_0): subscription. They're just like they use it when they need it, uh and and etc. And they and and especially for investing, they are not using it at all, or or very, very few people are doing it. And to to go back to your question, it's very important when you're using these tools is that you're learning. So the mindset
- 51:23
- Unidentified speaker (speaker_0): should be I'm using this to learn faster. I'm not using it to create a shortcut.
- 51:30
- Unidentified speaker (UNKNOWN): Now
- 51:30
- Unidentified speaker (speaker_0): when I explain the thesis tracker, it can seem like a shortcut because I'm like saving a lot of time because I don't have to look up all the information and structure it and then analyze the information. There is a pre-analysis. But why can I do that? Because over the 17 years I have acquired some skill
- 51:50
- Unidentified speaker (speaker_0): in investing to know what good looks like. And so when the system gives me information, I usually immediately know when it's off or when it's good.
- 52:02
- Unidentified speaker (UNKNOWN): And
- 52:02
- Unidentified speaker (speaker_0): it's the same thing in software development. They they call these evals. So you need to be able to evaluate the system based on its quality of the output. So my recommendation is always to go in with the mindset, especially if you're you're starting your investing journey, is with the goal should always be to
- 52:22
- Unidentified speaker (speaker_0): learn. And the the biggest advantage with with like an AI tool
- 52:26
- Unidentified speaker (UNKNOWN): like
- 52:26
- Unidentified speaker (speaker_0): Claude. is you can learn a lot faster than ever before. But if you're unsure what quality or what what what what what a good output looks like, then you have to be very careful.
- 52:37
- Unidentified speaker (speaker_3): Mm-hmm.
- 52:38
- Unidentified speaker (speaker_3): Yeah. Yeah. I think this is like
- 52:42
- Unidentified speaker (speaker_3): a very, very powerful lever for people that already have done the work to, as you mentioned understand what's their process, what does a good investment look like for me? Because at the end of the day, it's always personal and everyone has their own standards, right? So you'll have to
- 53:02
- Unidentified speaker (speaker_3): iterate with Claude and actually understand what's what's good and what's bad, so he also understands what's what's good It's and what's it's bad. And that makes a lot of sense. And it's easy to fall into like the shortcut mindset.
- 53:15
- Unidentified speaker (speaker_0): Yeah, but just to emphasize one thing, for example, a couple of weeks ago, Opus five point five came
- 53:21
- Unidentified speaker (UNKNOWN): out. And
- 53:22
- Unidentified speaker (speaker_0): I tested it. So what I usually do is when a new model comes out, I I do a test like like with Excel or something, like to to give it a DCF or evaluation and see if it makes any errors.
- 53:34
- Unidentified speaker (UNKNOWN): And
- 53:34
- Unidentified speaker (speaker_0): I do have to emphasize that this was the first time that I can give it the valuation a valuation mode. model or ask it to create based on a certain prompt a valuation on a certain company. And it did it in one go.
- 53:48
- Unidentified speaker (UNKNOWN): And
- 53:48
- Unidentified speaker (speaker_0): I checked the quality and there were no errors in
- 53:50
- Unidentified speaker (UNKNOWN): it.
- 53:51
- Unidentified speaker (speaker_0): So when I say avoid any shortcuts, of course, this is also a shortcut.
- 53:58
- Unidentified speaker (speaker_0): But it's a shortcut in a negative sense. If you if you've never created a DCF before because you have no idea what it means. So
- 54:07
- Unidentified speaker (speaker_0): if you know how a DCF works and and what each assumption means, then of course you can use Claude, do it in in like one minute,
- 54:16
- Unidentified speaker (UNKNOWN): and
- 54:16
- Unidentified speaker (speaker_0): it give you a it gives you a full DCF. If you've never created a DCF before, you will have to go through the steps. When I started my investing journey, I also uh followed the the YouTube videos for Professor De Moderan on valuation and I did all the steps. Now today do I still use an elaborate
- 54:36
- Unidentified speaker (speaker_0): Excel like uh De Moderin uses? No. My valuation work has become a lot simpler. But I did the journey. I did the steps to understand And why why do you take each step? And that's that's what you need to do. There is no shortcut for that. You need to learn.
- 54:54
- Unidentified speaker (speaker_3): Yep, and that that actually tucks into my other question, which is
- 55:01
- Unidentified speaker (speaker_3): what if you are starting out and you don't already have those like fundamentals? So you're starting to build the house from the roof.
- 55:11
- Unidentified speaker (UNKNOWN): Yeah.
- 55:12
- Unidentified speaker (speaker_3): As an analogy. Because nowadays, like what I imagine is like I I went through school for at least most part of it. I didn't have an AI to write to write my essays. So I actually had to learn like the grammatical cohesion of phrases and how things worked and how do you actually write. And so I
- 55:32
- Unidentified speaker (speaker_3): learned to write, and then I can sort of understand how why Claude is writing this way. or why does it look good or why why is it correct grammatically and stuff like that. And so to compare that to investing, if you've never as you mentioned, if you've never done a DCF or do you don't actually understand the the
- 55:51
- Unidentified speaker (speaker_3): reasoning behind it, uh you'll probably be uh very prone to making mistakes or at least not spotting mistakes from from an AI model. Right? So my question is what do you recommend people that are just starting out that maybe already have some basic knowledge are curious about investing
- 56:11
- Unidentified speaker (speaker_3): and are tempted to sort of build a full system of AI that will help them become the next Ron Buffett? What what would you recommend them to do first before uh starting to build all of these systems?
- 56:25
- Unidentified speaker (UNKNOWN): Yeah,
- 56:26
- Unidentified speaker (speaker_0): I have another a nephew who is exactly in that situation. Okay. He he's like twenty, twenty-one. He's studying economics at university, and he's now like because if you study economics at university, it's not the same as like a stock picking. And so he asked me to help him in in in how to learn all of this.
- 56:46
- Unidentified speaker (speaker_0): And
- 56:47
- Unidentified speaker (UNKNOWN): the
- 56:47
- Unidentified speaker (speaker_0): first thing I told him is he now has
- 56:51
- Unidentified speaker (speaker_0): basically a tutor next to him, which is called Claude, and he never sleeps. And you can ask him any question. And so, for example, he can learn accounting at university.
- 57:05
- Unidentified speaker (UNKNOWN): What
- 57:06
- Unidentified speaker (speaker_0): I, for example, did for him was I created a 10k reader app. So it's like uh you load in uh a 10k of a company.
- 57:15
- Unidentified speaker (UNKNOWN): and
- 57:15
- Unidentified speaker (speaker_0): then it will teach you how to read a 10k
- 57:19
- Unidentified speaker (speaker_0): based on accounting principles with practical examples of a real-world example, and he just learns a lot faster that way.
- 57:28
- Unidentified speaker (speaker_3): And
- 57:28
- Unidentified speaker (speaker_0): so, of course, he's not ready to build any like automated pipeline. He's now learning how to understand the basics and the fundamentals of a company. but he learns it a lot faster than I did in the past. Because he's using all these tools to learn. And it it that always should be the
- 57:48
- Unidentified speaker (speaker_0): first step.
- 57:49
- Unidentified speaker (speaker_3): I think that's extremely uh important as well, and probably good advice. And I again like investing and especially your style of investing is made with a lot of discipline, and I guess the that if you have if you want to have the discipline to be a good investor you also have to
- 58:09
- Unidentified speaker (speaker_3): have be disciplined enough to not like go straight into asking Claude Okay make me evaluation of this without actually understanding and so you have to go through the a certain process of learning the fundamentals and and to your point right now is probably the best time to learn anything uh uh ever. So
- 58:27
- Unidentified speaker (UNKNOWN): Yeah,
- 58:27
- Unidentified speaker (speaker_0): and I just want to emphasize that especially for young people, um there it's very difficult to create a shortcut. And why why do I why do I explain this? Is
- 58:39
- Unidentified speaker (speaker_0): investing is psychology and it's about conviction. And you need to build conviction in a company, and you cannot build conviction based on what Claude gives you. Because, for example, you ask Claude, Hey, do me a full analysis, you read the analysis, you think it's okay, and then you buy the company. That's not how it works. You
- 58:59
- Unidentified speaker (speaker_0): you use these AI tools and Claude, like early in the funnel, for example, to screen through a lot of ideas, and it goes a lot faster. Once you'd select a company you want to where you're more interested in and you want to dive deeper, you need to put in the work. There is no shortcut for that, but one of the shortcuts is okay,
- 59:19
- Unidentified speaker (speaker_0): I can let create my research agent creates a research report for me, and my contrarian agent creates a pushback report for me, but I still need to read both of them
- 59:29
- Ian Cassel: and
- 59:30
- Unidentified speaker (speaker_0): I need to understand them. And only when I have conviction then I will be able to go through the volatility of the stock in the future.
- 59:37
- Unidentified speaker (UNKNOWN): And
- 59:38
- Unidentified speaker (speaker_0): of course it a thesis tracker afterwards makes my life easier. But before buying the stock, I need to do a lot of work. And if young people who start their investing journey think that oh now they don't need to do that, there could be trouble in the future.
- 59:57
- Unidentified speaker (speaker_0): Exception is if you are a quant trading with an automated system, then it's different. But as long as you have to push the button to buy, sell at or trim, it's psychology. So
- 1:00:09
- Unidentified speaker (speaker_3): yeah, yeah. And I I don't think you can place trades with Cloud, but there's probably someone that made the bypass and uh
- 1:00:18
- Unidentified speaker (UNKNOWN): Yeah,
- 1:00:19
- Unidentified speaker (speaker_0): I know. I think I haven't tried it yet, but there is maybe a connector where you can push the trades, but I don't know if it's like a like like h live live trade. trading or or or fast trading. I'm I'm not sure about that. Yeah.
- 1:00:30
- Unidentified speaker (speaker_3): Yeah. Probably the the very sophisticated quant shops out there have already
- 1:00:36
- Unidentified speaker (speaker_3): sat. So uh it's probably not some retail guy doing it at uh the best scale. Perfect. So let me just ask you uh a wrap-up wrap up question because I I think we've went gone through a lot here someone that has gone through the whole episode is very excited
- 1:00:56
- Unidentified speaker (speaker_3): to implement some of the things you shared where would you start
- 1:01:00
- Unidentified speaker (speaker_0): so yeah the ba to to learn all of this
- 1:01:04
- Unidentified speaker (UNKNOWN): yeah
- 1:01:04
- Unidentified speaker (speaker_3): I'd say like uh let's say I'm I'm yeah I just want to implement your your workflow or like the idea the basic idea.
- 1:01:13
- Unidentified speaker (speaker_0): Yeah so if you want to see if you want to delve deeper in what I explained, of course it's a bit based on my process. I have a free newsletter. It's InvestWithAI Substack. It's just a Substack.
- 1:01:24
- Unidentified speaker (speaker_0): But I also have at investwithai.co
- 1:01:29
- Unidentified speaker (speaker_0): that's a full video course with the entire skill library. So
- 1:01:33
- Unidentified speaker (UNKNOWN): if
- 1:01:34
- Unidentified speaker (speaker_0): you say the skills, some of the skills uh that are presented in the slides, if you if you want to start from the beginning, because in my video course, like the the very first video is how to install cloth. Now somebody told me why did you make a video on how to install cloth? It's because of the settings. The first settings
- 1:01:53
- Unidentified speaker (speaker_0): you you you set before. Just to give you a a stupid example, if you install clots, cloth, I for example ID I deactivate the fact that it can learn from my knowledge, and that's important.
- 1:02:07
- Unidentified speaker (speaker_3): Okay.
- 1:02:08
- Unidentified speaker (speaker_0): So so yeah, invest with AI. There is all my all information. There is a free part, there is a paid part, but that's up to you uh to
- 1:02:17
- Unidentified speaker (UNKNOWN): to
- 1:02:17
- Unidentified speaker (speaker_0): find all the information.
- 1:02:19
- Unidentified speaker (UNKNOWN): If
- 1:02:19
- Unidentified speaker (speaker_2): you enjoyed this conversation, I'm sure you'll love microcapclow.com. microcapclub.com. And subscribe if you'd like to follow along. Thank you so much. for listening.
More from MicroCapClub
Ex-intelligence officer: how to tell when a CEO is lying | Gorm Thomassen
Episode discussion: Ex-intelligence officer Gorm Thomassen shares techniques for detecting deception and evaluating management quality
Speakers’ claims: Gorm Thomassen, co-founder of AKO Capital, discusses his background in Norwegian military intelligence and how he applies interrogation techniques to equity research. He emphasizes distinguishing between information and persuasion, using specific heuristics to filter out marketing fluff from verifiable data. Thomassen outlines AKO's internal processes, including bias detection, ecosystem analysis, and accounting scoring, to improve forecast accuracy. He also shares practical interviewing strategies, such as using silence and open-ended questions, to detect evasive language and uncover true management sentiment.
Find the Bull Market by Ian Cassel
Episode discussion: Ian Cassel argues investors should find individual bull markets within bear markets by targeting specific sectors and stocks.
Speakers’ claims: Ian Cassel uses historical examples to argue that broad market indices mask underlying growth opportunities during downturns. He cites Jesse Livermore's 1929 short, Homestake Mining's rise during the Great Depression, and Tractor Supply's performance in the dot-com crash as evidence that specific stocks can thrive when indices fall. Cassel suggests that optimism drives bull markets while fear drives bear markets, creating distinct sectors that benefit from each environment. He concludes that investors should focus on identifying these niche bull markets rather than making excuses for market downturns.