WEBVTT

1
00:00:10.000 --> 00:00:37.627
Welcome to the Global Advisors A.I Voices podcast. I'm Maya.
Today is Thursday, the twenty-ninth of October 2026.
Your morning catch-up on A.I. Our goal is to help you feel like you spent the last twenty four hours in a Silicon Valley coffee shop, with an ear on China and the wider world.
We bring together the conversations, announcements and debates that matter, with the original voices clearly attributed. Please like and follow the show for your daily catch-up.

2
00:00:39.676 --> 00:00:47.292
Google DeepMind has announced Genie 4, a new world model designed to generate interactive simulations in real time.

3
00:00:47.642 --> 00:00:55.172
Unlike standard video generators that produce fixed clips, the system builds environments frame by frame as new inputs arrive.

4
00:00:55.522 --> 00:01:04.866
The model is natively multimodal, generating visual scenes alongside ambient noise, music, and interactive dialogue simultaneously.

5
00:01:05.216 --> 00:01:15.627
To keep simulated spaces coherent, DeepMind uses Gemini models to anchor appearance and physical behavior while steering multi-step tasks using text prompts.

6
00:01:15.977 --> 00:01:26.985
For now, the company describes Genie 4 as an early experimental research preview, rolling out first to trusted testers across gaming, engineering, and creative industries.

7
00:01:27.335 --> 00:01:41.628
Secondary press briefings note practical boundaries: while scene consistency reportedly reaches up to two minutes, interactive sessions remain limited to a few minutes, and simulated objects begin to drift after repeated prompt changes.

8
00:01:44.028 --> 00:01:53.649
DeepMind has also introduced another frontier model, Gemini 4 Carbon, with an initial focus on long-horizon coding and cybersecurity tasks.

9
00:01:53.999 --> 00:02:03.684
Leadership at Google DeepMind characterizes Carbon as its strongest model to date, highlighting its ability to handle complex, sustained assignments.

10
00:02:04.034 --> 00:02:14.402
You cannot simply sign up and run it, however. Carbon lacks a publicly accessible programming interface and remains unavailable outside Google's controlled access programs.

11
00:02:14.752 --> 00:02:25.483
Google is pacing deployment through pre-release evaluations with the United States government and initial access for vetted cyber defenders through its Fairwind Program.

12
00:02:25.833 --> 00:02:36.500
Google says Carbon will match the introductory rates of its existing Argon model once it launches, though general availability for businesses and consumers remains undated.

13
00:02:38.899 --> 00:02:54.536
Frontier laboratories are also adjusting how they disclose complex theoretical findings. OpenAI announced that its internal models produced scientific results across pure mathematics, computational complexity, and theoretical physics.

14
00:02:54.886 --> 00:03:09.990
In mathematics, the company says its models generalized a nineteen ninety-one wave conjecture across all spatial dimensions, while in computer science, it reported a claimed disproof of the long-standing Exponential Time Hypothesis.

15
00:03:10.340 --> 00:03:18.852
In physics, the company cited partial progress toward an existence proof for certain interacting quantum field theories in four dimensions.

16
00:03:19.202 --> 00:03:41.815
Rather than releasing all these claims at once, OpenAI used distinct disclosure paths. After past releases caused verification bottlenecks among academic researchers, the lab published the mathematical papers publicly but coordinated with external experts to evaluate the computer science findings, while deferring the release of incomplete physics materials.

17
00:03:42.165 --> 00:03:57.291
OpenAI says it will not seek academic credit for the discoveries, leaving attribution to the collaborating human authors. Observers note that the computer science claim remains strictly contingent on passing formal academic peer review.

18
00:03:59.690 --> 00:04:12.554
In corporate finance, Anthropic has publicly filed its Form S-1 registration statement with the SEC for a proposed initial public offering on the Nasdaq under the proposed symbol ANT.

19
00:04:12.904 --> 00:04:22.205
The preliminary prospectus leaves the offering share count, price range, and valuation terms blank, but it outlines an unusual governance structure.

20
00:04:22.555 --> 00:04:34.011
Under the proposal, Anthropic's seven co-founders will retain fifty-point-one percent of total shareholder voting power through a single share of Class F stock held via an LLC.

21
00:04:34.361 --> 00:04:51.919
That single share carries no economic rights or financial claims. It is designed so that incoming public Class A shares are capped at forty-nine-point-nine percent voting power, leaving charter amendments, corporate sales, and non-trust board elections firmly under founder control.

22
00:04:54.318 --> 00:05:00.270
That same registration filing offers a window into the velocity of commercial A.I adoption.

23
00:05:00.620 --> 00:05:16.001
Third-party analysis of the disclosure notes that more than eighty percent of Anthropic's revenue in the September quarter came from model versions released within the previous six months, illustrating that customers migrate quickly to the newest capabilities.

24
00:05:16.351 --> 00:05:30.346
The commentary highlights a distribution strategy built across cloud partner marketplaces, first-party tools, and third-party software integrations to reach enterprise clients without requiring entire software overhauls.

25
00:05:30.696 --> 00:05:46.674
The filing reports more than half of the Fortune 500 paying Anthropic at least one hundred thousand dollars annually, though analysts emphasize that long-term expansion depends on doing the slow work of embedding models into core operational workflows.

26
00:05:49.074 --> 00:05:56.583
Looking more closely at those enterprise accounts reveals a striking gap between customer reach and customer spending depth.

27
00:05:56.933 --> 00:06:12.357
In the September quarter, following sixteen-point-three billion dollars in first-half revenue, publicly traded Fortune 500 customers spent an average of less than one dollar with Anthropic for every one thousand dollars in broader operating expenses.

28
00:06:12.707 --> 00:06:23.877
Commentators argue this low budget penetration suggests corporate A.I deployments remain in early departmental pilot stages rather than defining the true size of the market.

29
00:06:24.227 --> 00:06:42.851
At the same time, investors examining the prospectus must weigh Anthropic's rapid top-line growth against significant net losses, large long-term compute commitments, and notable customer concentration, with five clients generating roughly a quarter of revenue in the first nine months of the year.

30
00:06:45.250 --> 00:06:56.599
Meanwhile, major hardware and cloud partners continue to tighten commercial and financial ties. Nvidia has agreed to invest three billion dollars in Meta's equity offering.

31
00:06:56.949 --> 00:07:06.826
That commitment is part of Meta's wider capital plan to raise ninety-five billion dollars, beginning with an initial thirty-five billion dollar public offering.

32
00:07:07.176 --> 00:07:22.771
Alongside the investment, Meta plans to expand its deployment of Nvidia Vera Rubin systems, central processors, and high-throughput networking equipment, while the companies explore joint work on monitoring the behavior of autonomous software agents.

33
00:07:23.121 --> 00:07:35.281
The transaction continues a sharp acceleration in direct dealmaking across the artificial intelligence sector, with Nvidia participating in over one hundred funding rounds this year alone.

34
00:07:37.680 --> 00:07:55.152
Outside corporate balance sheets, artificial intelligence is testing traditional legal procedures. Geoffrey Vos, the retiring Master of the Rolls and head of England's civil justice system, is calling for public debate on whether judges should use A.I to help resolve disputes.

35
00:07:55.502 --> 00:08:06.254
Current judicial guidance in England advises judges to restrict machine assistance to administrative duties such as finding documents and summarizing dense texts.

36
00:08:06.604 --> 00:08:17.356
Vos pointed to courts in Shenzhen, China, where judges already decide civil cases with substantial machine assistance, arguing that courts worldwide should examine the model.

37
00:08:17.706 --> 00:08:32.832
He suggested that while complex commercial disputes and sensitive family matters must remain with human judges, routine cases like personal injury or low-value negligence claims could be resolved with machine assistance to lower costs.

38
00:08:33.182 --> 00:08:50.803
Vos acknowledged that the road carries serious pitfalls, from lawyers citing fabricated precedents to a recent case of a United Kingdom lawyer being struck off for misusing the technology, emphasizing that society must decide where to draw the boundary for machine-assisted justice.

39
00:08:53.202 --> 00:09:14.514
Finally, new workforce research suggests that having the same tools and training does not produce identical results. A study by KPMG and the University of Texas at Austin examining more than five hundred early-career professionals found that baseline skill equivalence did not guarantee similar performance when using A.I.

40
00:09:14.864 --> 00:09:27.408
The researchers labeled top performers as A.I amplifiers, reporting that their advantage came from how they guided, evaluated, and refined model outputs rather than from specialized knowledge alone.

41
00:09:27.758 --> 00:09:46.297
While the finding comes from a broadcast sponsorship context, it matches a wider enterprise challenge: as frontier tools like Claude generate ninety percent of developer code at specialized labs, analysts note a massive divide remains between aggressive early adopters and the broader corporate workforce.

42
00:09:46.647 --> 00:10:03.095
That's A.I Voices for today. I'm Maya. Thanks for listening.
Follow the show on Spotify, YouTube, Apple Podcasts, Amazon Music, or your favourite podcast platform, and explore the original voices and sources in the episode notes. See you tomorrow.