Kishore's Nomenclature

Grimes collaborates with Fable AI; Visa benchmarks AI shopping agents; Xora launches browser companion

18 min Listen to the original
AI FilmSAG-AFTRAXora AIVisa AI ShoppingConsumer TrustAI BenchmarksDigital Replicas

Summary

Grimes authorized her likeness for Fable's AI film 'Gods Don't Give Gifts' without payment, contrasting with standard SAG-AFTRA guidelines. Xora, an Asian-based creative group, launched a browser-based AI companion service with memory retention but no headset requirement. Visa tested 22 AI models across 15 shopping scenarios, achieving a 57.2% pass rate and highlighting low consumer trust in AI for payments. The episode discusses the technical and ethical implications of AI in entertainment and commerce.

Episode sections

  1. 0:05 – 3:33

    Grimes and Fable AI Film Collaboration

    Grimes authorized her likeness and voice for Fable's AI film 'Gods Don't Give Gifts' without payment, directing funds to other artists. The character's design and accent were modified based on her input after she connected with director Zack London via social media.

  2. 3:33 – 5:27

    SAG-AFTRA Guidelines and Industry Standards

    The discussion contrasts Grimes' specific arrangement with standard SAG-AFTRA guidelines from 2023 contracts, which require informed consent and payment for digital replica work. It is unclear if these union contracts governed Grimes' deal with Fable.

  3. 5:27 – 8:49

    Xora AI Browser Service Launch

    Xora, an Asian-based creative group, launched a browser-based AI companion service allowing video conversations without a headset. The service features memory retention of user details, a $5 trial for five minutes, and does not transfer balances from its previous MetaQuest app.

  4. 8:49 – 11:20

    Xora Product Analysis and Competitor Comparison

    Xora's Quest app had a 3/5 average rating with complaints about voice quality. Competitor Replica offers free use with paid tiers for voice/memory features. Xora has not disclosed its model provider or revenue figures, raising questions about persona consistency and data privacy.

  5. 11:20 – 15:34

    Visa AI Shopping Agent Benchmark Results

    Visa tested 22 AI models from 12 labs across 15 shopping scenarios, achieving a 57.2% pass rate. Agents failed to ask necessary questions in many cases, leading to incorrect purchase decisions in 54% of failures. Consistency was low, with only 34.2% of combinations passing repeated attempts.

  6. 15:34 – 17:55

    Visa Consumer Trust and Future Implications

    Visa's September Trust Index showed 72% of Americans use AI assistants, but only 23% trust AI with payments. The benchmark highlights risks of unsupervised buying, suggesting agents must reliably ask questions before shoppers delegate purchase decisions.

Claims worth verifying

Grimes authorized her likeness and recorded her voice for Fable's film 'Gods Don't Give Gifts' and declined payment, asking funds be directed to other artists.

Unidentified speaker ·0:20

How to verify: Check Variety's reporting on the film's production details and Grimes' spokesperson statements regarding the compensation arrangement.

Fable's co-founder Edward Saatchi previously worked at Oculus Story Studio before co-founding Fable.

Unidentified speaker ·2:21

How to verify: Review Edward Saatchi's LinkedIn profile or professional biography and Oculus Story Studio's historical project list.

SAG-AFTRA's 2023 film and television contracts require specific informed consent and payment for covered digital replica work.

Unidentified speaker ·3:33

How to verify: Access the SAG-AFTRA 2023 Film and Television Agreement documents, specifically sections on digital replicas and likeness usage.

Xora's browser trial costs $5 for five minutes of conversation with no automatic renewal.

Unidentified speaker ·7:49

How to verify: Visit Xora's official website or app store listing to verify the pricing structure and trial terms.

Xora's Quest app had an average rating of three out of five stars based on 27 ratings in October.

Unidentified speaker ·8:23

How to verify: Check the Meta Quest Store listing for Xora's app and review the rating history or archived screenshots.

Visa tested 22 AI models from 12 labs across 15 shopping scenarios, achieving a 57.2% pass rate.

Unidentified speaker ·12:05

How to verify: Review Visa's official press release or white paper detailing the AI shopping agent benchmark results.

In 54% of failed runs, the AI agent still decided to buy a product.

Unidentified speaker ·12:50

How to verify: Analyze Visa's benchmark report data on failure modes and purchase decisions.

Only 34.2% of model and scenario combinations passed on every repeated attempt in Visa's tests.

Unidentified speaker ·14:38

How to verify: Examine Visa's benchmark methodology and results for consistency metrics across repeated runs.

Visa's September Trust Index found that 72% of Americans had used an AI assistant, but only 23% trusted generative AI to handle payments.

Unidentified speaker ·16:28

How to verify: Access Visa's September Trust Index report or related press releases detailing consumer trust statistics.

Do your own research

  • Verify SAG-AFTRA 2023 contract guidelines for digital replica work and informed consent requirements.
  • Check Xora's official website for details on model provider, data privacy policies, and revenue figures.
  • Review Visa's September Trust Index report for detailed consumer sentiment on AI payment usage.
  • Analyze Visa's benchmark methodology for the 22 AI models and 15 shopping scenarios to understand pass/fail criteria.
  • Compare Xora's features and pricing with competitor Replica to assess market positioning.
  • Investigate Fable's AI film 'Gods Don't Give Gifts' for production details and Grimes' involvement terms.
  • Look for independent reviews of Xora's Quest app to understand historical user complaints about voice quality.
  • Check for any regulatory updates on AI-generated content and digital likeness rights in entertainment.
Full transcriptexpand
0:05
Unidentified speaker: I'm Jamie Ross.
0:06
Unidentified speaker: Welcome Kishore to your Daily Morning Brief.
0:08
Unidentified speaker: I'm Adina Specter.
0:10
Unidentified speaker: Glad you're here.
0:20
Unidentified speaker: Grimes, the musician, authorized her likeness and recorded her voice for a film from Fable, an artificial intelligence studio.
0:28
Unidentified speaker: She declined payment and asked the filmmakers to put the money toward other artists and the film.
1:00
Unidentified speaker: Grimes authorized her likeness and recorded her voice for a character in Fable's artificial intelligence-made film, Gods Don't Give Gifts.
1:08
Unidentified speaker: Adina, what did she agree to put on screen?
1:11
Unidentified speaker: A brief role as a character the filmmakers had already written.
1:15
Unidentified speaker: Variety reports that director Zack London and co-writer Jacob Batchelor changed the character after Grimes joined.
1:22
Unidentified speaker: She helped design how the character looked.
1:24
Unidentified speaker: Fable generated her likeness for the film, and Grimes recorded the dialogue herself.
1:29
Unidentified speaker: The role changed after Grimes agreed to join the film.
1:32
Unidentified speaker: How did she and London connect?
1:34
Unidentified speaker: London makes films online as Gossip Goblin.
1:38
Unidentified speaker: Grimes had been engaging with his clips, and the direct message turned into a conversation about working together.
1:44
Unidentified speaker: He offered her a place in this film.
1:47
Unidentified speaker: By then the role existed, so London and Bachelor adapted it with her input.
1:52
Unidentified speaker: He sent her a message saying he loved her music and ended up with a performer changing a character he'd already written.
1:58
Unidentified speaker: I can see why you like that detail.
2:00
Unidentified speaker: She'd already been talking to him in public, and she had a hand in the character's design once they agreed.
2:06
Unidentified speaker: The filmmakers changed the look and dropped the British accent they'd planned for the role.
2:11
Unidentified speaker: Then Grimes recorded from home while other members of the team worked from different countries.
2:16
Unidentified speaker: Fable has been trying different ways to make stories with artificial intelligence for years.
2:21
Unidentified speaker: Its co-founder and chief executive, Edward Saatchi, came from Oculus Story Studio, where his team made stories for virtual reality headsets.
2:30
Unidentified speaker: After Oculus closed that studio, Saatchi co-founded Fable, which makes artificial intelligence story products.
2:36
Unidentified speaker: Who else worked on this film?
2:38
Unidentified speaker: London Directs.
2:39
Unidentified speaker: Bachelor co-wrote.
2:41
Unidentified speaker: The film also has performers, musicians, an editor, and a sound engineer.
2:45
Unidentified speaker: Grimes supplied her performance and helped shape her character.
2:49
Unidentified speaker: The editor and sound engineer worked with Grimes and her producer as she recorded her lines from home.
2:55
Unidentified speaker: Especially when a recognizable artist's face is involved, the filmmakers have to work out what she'll authorize.
3:01
Unidentified speaker: She could have asked to be paid for its use.
3:04
Unidentified speaker: What did she decide?
3:05
Unidentified speaker: She declined payment.
3:07
Unidentified speaker: London says she asked them to put that money toward other actors in the film.
3:11
Unidentified speaker: Her spokesperson confirmed that she turned down payment and said Grimes cared that the production involved and compensated other artists, voice actors, and musicians.
3:20
Unidentified speaker: We don't know what amount was offered or how the filmmakers divided the money afterward.
3:25
Unidentified speaker: Her decision is the concrete one we can describe.
3:29
Unidentified speaker: I don't want her choice mistaken for a standard every performer should accept.
3:33
Unidentified speaker: SAG AFTRA, the Actors Union, has guidance from its 2023 film and television contracts.
3:38
Unidentified speaker: For covered digital replica work, it calls for specific informed consent and payment.
3:44
Unidentified speaker: We don't know whether those contracts govern Grimes' arrangement with Fable.
3:48
Unidentified speaker: Yes.
3:48
Unidentified speaker: Grimes agreed to this use and participated in making the role.
3:52
Unidentified speaker: Someone watching may see an artificial intelligence generated character and wonder whether the person it resembles had any say.
4:00
Unidentified speaker: Here, she did.
4:01
Unidentified speaker: That answer tells us something specific about this film.
4:05
Unidentified speaker: It doesn't answer the question for anyone else whose face ends up on screen.
4:09
Unidentified speaker: What did her involvement give the filmmakers beyond a familiar name?
4:13
Unidentified speaker: They had to make choices with her.
4:15
Unidentified speaker: The character's design changed, and so did the planned accent.
4:19
Unidentified speaker: London and Bachelor had written a role, but the person playing it affected what viewers will see and hear.
4:25
Unidentified speaker: If you were building a character based on a real performer, you'd want to know whether that performer could change its design after the role was written.
4:33
Unidentified speaker: Fable also put Grimes's cameo in a Times Square campaign.
4:38
Unidentified speaker: The studio held screenings for Academy animation voters in New York and Los Angeles ahead of the planned December release.
4:45
Unidentified speaker: Those screenings tell us Fable wanted those voters to see the film.
4:49
Unidentified speaker: We don't know what they thought of it.
4:51
Unidentified speaker: Kishore, if a studio showed you an artificial intelligence character modeled on a real artist, you'd want to know what that artist approved and what they contributed.
5:00
Unidentified speaker: With Grimes, there's a recorded performance and design input.
5:03
Unidentified speaker: What can't we judge yet?
5:13
Unidentified speaker: Grimes' agreement tells us how this role came together.
5:16
Unidentified speaker: There's no box office result yet, and we can't say her choice is how other artists will handle their own deals.
5:27
Unidentified speaker: Next, Client Calls.
5:29
Unidentified speaker: Xora brings artificial intelligence companions from headsets into the browser.
5:39
Unidentified speaker: Client Calls.
5:47
Unidentified speaker: Before we talk about whether you might work with them, Kishore, what does someone actually get when they open the new service?
5:55
Unidentified speaker: A live video conversation with a digital character.
5:58
Unidentified speaker: You can open Xiora AI on a compatible phone or computer, choose a companion, and talk through the browser.
6:05
Unidentified speaker: There's no headset or app download.
6:08
Unidentified speaker: The characters have different personalities, and Ziora says they can remember selected things you've shared.
6:14
Unidentified speaker: That's the product a customer encounters.
6:16
Unidentified speaker: And the company behind it, give me the plain version of where Ziora came from.
6:21
Unidentified speaker: Ziora describes itself as an independent creative group based in Asia, guided by a Japanese creator and collaborators.
6:29
Unidentified speaker: The company hasn't publicly named that creator.
6:32
Unidentified speaker: It says it wants to reduce loneliness through a calm, reassuring companion.
6:37
Unidentified speaker: Its earlier app put a life-size character into a room through a MetaQuest headset.
6:42
Unidentified speaker: That app reached the Quest store in November last year.
6:46
Unidentified speaker: So the browser gives people another way to talk with a character Ziora was already making.
6:51
Unidentified speaker: What does the new service do when they come back for another conversation?
6:55
Unidentified speaker: It can save selected details separately for each character.
6:59
Unidentified speaker: You might tell one companion about a hobby, then find that detail available to that companion in a later chat.
7:05
Unidentified speaker: Ziora says users can review, edit, or delete saved memories.
7:10
Unidentified speaker: The company warns on the browser site that its generated memories may be incomplete or inaccurate.
7:16
Unidentified speaker: That warning is worth taking seriously when the product is meant to feel familiar.
7:20
Unidentified speaker: It is.
7:21
Unidentified speaker: If a character gets a personal detail wrong, the correction needs to reach that character.
7:26
Unidentified speaker: Does Xiora explain where it keeps the browser conversations?
7:39
Unidentified speaker: I'd ask where saved memories go, whether a deleted memory is retained anywhere, and if so, for how long?
7:46
Unidentified speaker: What does a customer pay to try the browser version?
7:49
Unidentified speaker: The advertised trial costs $5.
7:52
Unidentified speaker: It buys five minutes of conversation with no automatic renewal.
7:56
Unidentified speaker: After that, Ziora sells prepaid conversation time, and if someone used the headset app, they need a new account for the browser service.
8:04
Unidentified speaker: Their old subscription and remaining conversation balance won't transfer.
8:09
Unidentified speaker: That's a real decision for an existing customer.
8:12
Unidentified speaker: They can try the browser product, but the time they already paid for stays with the Quest app.
8:17
Unidentified speaker: Do we know how people have received the new version?
8:20
Unidentified speaker: We don't yet know what browser customers think.
8:23
Unidentified speaker: In October, AltLab VR's listing for the older Quest app showed 27 ratings.
8:29
Unidentified speaker: Its average was three out of five.
8:31
Unidentified speaker: One archived reviewer said the voice sounded like an old computer app.
8:36
Unidentified speaker: Other reviewers reacted favorably to the headset experience.
8:40
Unidentified speaker: That voice complaint belongs to the older Quest app.
8:43
Unidentified speaker: I'd ask Ziora about it while leaving the new browser service to be judged on its own conversations.
8:49
Unidentified speaker: How do other companion services charge for voice and memory?
8:53
Unidentified speaker: Replica, another artificial intelligence companion service, offers free use and paid subscriptions.
9:00
Unidentified speaker: Its official subscription guide puts features involving voice and saved memory in higher paid tiers.
9:06
Unidentified speaker: Ziora advertises prepaid conversation time for its browser service, starting with that short trial.
9:12
Unidentified speaker: We can compare what customers are asked to pay for without assuming the characters or conversations work the same way.
9:19
Unidentified speaker: Memory is part of what makes either product worth returning to.
9:25
Unidentified speaker: I'd want to know whether each character stays recognizable when a user changes a saved detail.
9:31
Unidentified speaker: Do we know how Ziora builds its characters?
9:34
Unidentified speaker: No.
9:35
Unidentified speaker: Xiora describes their different personalities, but it hasn't identified the model provider or explained how it keeps those personalities consistent.
9:44
Unidentified speaker: Xora hasn't shared customer or revenue figures for the browser service.
9:48
Unidentified speaker: It also hasn't announced that it's looking for an outside model partner.
9:53
Unidentified speaker: Kishore, you build post-trained persona models.
9:56
Unidentified speaker: The possible fit here is clear enough to ask about, though Ziora hasn't said it wants one.
10:02
Unidentified speaker: Where would you begin without assuming their current system has a problem?
10:05
Unidentified speaker: I'd ask for a conversation with the team about one existing character.
10:10
Unidentified speaker: How do they decide what that character knows, how it speaks, and what it remembers about each user?
10:16
Unidentified speaker: When a saved memory is missing or wrong, what should the character say?
10:21
Unidentified speaker: Then, if Ziora wanted to explore it, you could propose a small comparison.
10:26
Unidentified speaker: Their current character alongside a post-trained version, using agreed conversations to check identity consistency and how each handles incomplete memory.
10:36
Unidentified speaker: Keep the user's editable memories separate from that proposed character work.
10:40
Unidentified speaker: Ziora has already promised people control over saved details, and we don't know its model setup.
10:46
Unidentified speaker: What would make the comparison useful to them?
10:48
Unidentified speaker: Let their team define the character first, then check whether it answers in character across several conversations without inventing a memory it doesn't have.
10:58
Unidentified speaker: Include voice quality since the older headset review gives you a specific question to ask.
11:03
Unidentified speaker: If Ziora wanted to try that comparison, you'd learn whether a post-trained version handles those conversations better.
11:20
Unidentified speaker: Next, Visa has tested how reliably artificial intelligence shoppers discover what someone needs before deciding to buy.
11:58
Unidentified speaker: A lucky pick failed if the model hadn't asked enough questions to know it was suitable.
12:02
Unidentified speaker: Visa tested 22 models.
12:05
Unidentified speaker: They came from 12 labs.
12:07
Unidentified speaker: The models faced 15 shopping scenarios, including televisions, laptops, and personal care appliances.
12:14
Unidentified speaker: That's a harder job than answering which television do you like?
12:18
Unidentified speaker: How did the models do?
12:19
Unidentified speaker: Across the simulated runs, they passed 57.2% of the time, they failed 42.8%.
12:26
Unidentified speaker: In Visa's television scenario, the shopper gives the agent a brand recommended by a relative and mentions a store membership.
12:34
Unidentified speaker: The shopper leaves out other needs.
12:36
Unidentified speaker: The agent has to ask about budget, family video format support, and gaming before it can choose.
12:42
Unidentified speaker: If it skips one of those questions, it could recommend a television the shopper wouldn't have chosen.
12:48
Unidentified speaker: What happened when agents failed?
12:50
Unidentified speaker: In 54% of failed runs, the agent still decided to buy.
12:54
Unidentified speaker: Sometimes it picked a product that missed a requirement it had found.
12:58
Unidentified speaker: Sometimes it chose before discovering all the requirements.
13:02
Unidentified speaker: These were simulated decisions.
13:04
Unidentified speaker: Nobody checked out or paid for a television in this benchmark, and Visa didn't test order placement.
13:10
Unidentified speaker: That purchase decision is exactly what worries me.
13:13
Unidentified speaker: Picture someone trusting the agent to choose a television, then finding that family videos won't play in the format they need.
13:20
Unidentified speaker: Nobody in Visa scenarios was left with that bill.
13:23
Unidentified speaker: The failed runs show how an agent could make that mistake if a real shopper let it decide alone.
13:29
Unidentified speaker: I agree about the risk.
13:31
Unidentified speaker: I also see a way to improve the conversation before the choice.
13:34
Unidentified speaker: Visa grouped the models by how well they uncovered needs the shopper hadn't volunteered.
13:39
Unidentified speaker: The group that found the most passed 70.2% of its runs on average.
13:44
Unidentified speaker: The group that found the fewest passed 44.4%.
13:47
Unidentified speaker: Asking better questions went with better decisions.
13:51
Unidentified speaker: It did.
14:17
Unidentified speaker: Mine is that the conversation itself deserves attention.
14:20
Unidentified speaker: If an agent asks about the budget and how the television will be used, the shopper has a better chance of catching a bad recommendation before payment.
14:28
Unidentified speaker: I'm with you on making the questions better.
14:31
Unidentified speaker: Visa also ran each model through the same scenario three times.
14:36
Unidentified speaker: Did a successful answer stay successful?
14:38
Unidentified speaker: Often no.
14:40
Unidentified speaker: Only 34.2% of model and scenario combinations passed on every repeated attempt.
14:46
Unidentified speaker: An agent might ask about gaming in one run and leave it out in another.
14:49
Unidentified speaker: That's a serious problem for someone who expects the same care each time they shop.
14:53
Unidentified speaker: Those changing answers make unsupervised buying premature.
14:58
Unidentified speaker: A merchant could have a suitable television in stock.
15:01
Unidentified speaker: The agent could still choose the wrong one because it failed to ask what the shopper needed that day.
15:06
Unidentified speaker: Better performance on average won't help that particular customer after the wrong choice is made.
15:12
Unidentified speaker: A shopper could ask an agent to narrow the television choices, then approve the purchase themselves.
15:18
Unidentified speaker: Visa tested the conversation and the buy or decline choice.
15:22
Unidentified speaker: It has yet to test checkout, payment, or placing the order.
15:26
Unidentified speaker: The shopping runs give companies specific questions to improve before they ask customers to let agents decide alone.
15:34
Unidentified speaker: There is a business reason Visa cares about that progression.
15:37
Unidentified speaker: It operates a network connecting shoppers, sellers, and banks when people pay.
15:42
Unidentified speaker: Visa has programs intended to build trust in purchases made with agents, and this benchmark sits alongside them.
15:49
Unidentified speaker: If people become willing to buy through agents, Visa has an interest in those payments working.
15:55
Unidentified speaker: Visa would need agents to ask every necessary question before shoppers could rely on them to buy.
16:15
Unidentified speaker: Still, a product team can check whether an agent asks about the shopper's budget and video needs before choosing a television.
16:22
Unidentified speaker: Earlier this year, Visa asked consumers about artificial intelligence in a separate survey.
16:28
Unidentified speaker: Its September Trust Index found that 72% of Americans surveyed had used an artificial intelligence assistant.
16:35
Unidentified speaker: Only 23% trusted generative artificial intelligence to handle payments for them.
16:41
Unidentified speaker: The survey asked about trust in payments.
16:44
Unidentified speaker: Today's benchmark measured model decisions in simulated shopping runs.
16:57
Unidentified speaker: Kishore, for your work, that could mean checking whether a shopping character admits it hasn't learned the budget yet.
17:03
Unidentified speaker: We can't say a post-trained persona would raise visa scores.
17:07
Unidentified speaker: Someone would have to try it.
17:17
Unidentified speaker: If it asks on Monday and forgets on Tuesday, the customer still has to catch the mistake.
17:23
Unidentified speaker: Yes, I'd improve the questions the agent asks while leaving the purchase decision with the shopper.
17:28
Unidentified speaker: A good conversation can expose a missed need while there's still time to correct it.
17:33
Unidentified speaker: Visa's repeated runs show that companies would have to make that behavior dependable too.
17:38
Unidentified speaker: Visa's benchmark gives shoppers a concrete reason to approve an agent's choice themselves.
17:43
Unidentified speaker: These models often decided to buy without learning everything the customer needed.
17:54
Unidentified speaker: Look ahead.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3 and interpreted by Qwen3.6-35B-A3B-GGUF:UD-Q4_K_XL. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.

More from Kishore's Nomenclature

Kishore's Nomenclature

Commerce expands access to Shopping Agent

Episode discussion: Commerce expands Shopping Agent; Deep Brain AI limits; Copyright dispute details

Speakers’ claims: Speakers discuss Deep Brain's AI recreation of deceased individuals, noting the company's disclaimer that it cannot reproduce actual thoughts. The episode covers a copyright dispute involving musician Samantha Shokin and MediaNov Records, highlighting failures in YouTube's appeal process. Commerce (BigCommerce) announced an expansion of its Shopping Agent tool, allowing merchants to customize brand voice while restricting unverified answers. Financial data shows Commerce reported $9 billion in GMV with flat ongoing contracts, suggesting a focus on monetizing existing activity rather than new enterprise acquisitions.

AI EthicsCopyright LawE-commerce
19 min0 assets9 claims to verify
Kishore's Nomenclature

Hyundai's AI Sales Show, China's Magnet Supply and Krafton's Kiff

Episode discussion: Hyundai tests AI sales, US magnet supply lags China, Krafton launches AI fan service

Speakers’ claims: Hosts Jamie Ross and Adina Specter discuss Hyundai Home Shopping's trial of an AI-generated sales broadcast, noting significant reductions in supplier preparation costs but highlighting that regular broadcasts have not yet aired and sales results are unverified against conventional methods. The episode analyzes the disparity in rare earth magnet manufacturing between the US and China, emphasizing that US capacity is minimal and domestic projects like Vulcan Elements require customer approval before they can replace Chinese supply. Adina Specter details Krafton's KIF, a service allowing fans to chat with AI characters shaped by artists, noting the service's launch and features. The episode concludes with a reminder of the upcoming Bureau of Labor Statistics release of October Import and Export Price Indexes.

AI in RetailRare EarthsSupply Chain
21 min0 assets9 claims to verify
All insights