WEBVTT

1
00:00:10.000 --> 00:00:37.584
Welcome to the Global Advisors A.I Voices podcast. I'm Maya.
Today is Friday, the twenty-third 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
00:00:39.633 --> 00:00:59.708
We begin with the intensifying scramble for private balance sheets. Reporting from CNBC indicates that OpenAI is close to finalizing a funding round of up to thirty billion dollars at a pre-money valuation of one point two five trillion dollars, a figure that does not include the new money being brought in.

3
00:01:00.058 --> 00:01:11.727
The transaction is reportedly led by investment firm Coatue alongside investors from the United Arab Emirates, who are positioning the capital as a bridge ahead of an eventual public offering.

4
00:01:12.077 --> 00:01:31.768
Terms remain under active negotiation and could shift, with reports suggesting the deal could also tie into securing additional regional computing capacity. If completed, commanding a valuation north of a trillion dollars would break records as the highest valuation ever recorded for a private business.

5
00:01:32.118 --> 00:01:49.142
Inside OpenAI, leadership timing has diverged: while Chief Financial Officer Sarah Friar previously pointed to 2027 for a public debut, Chief Executive Sam Altman recently argued that right now would be an ill-advised moment to go public.

6
00:01:51.541 --> 00:02:09.930
That massive private round is taking place on a tight clock. Reports from Reuters indicate that OpenAI has accelerated its talks to close within weeks, attempting to lock in its financing before rival Anthropic moves ahead with plans for an initial public offering as soon as mid-November.

7
00:02:10.280 --> 00:02:31.763
Market observers speculate that OpenAI wants its valuation established before public investors get a clear look at Anthropic's audited financial statements. The two developers are competing aggressively for scarce capital against a backdrop of technology stock volatility and rising scrutiny over artificial intelligence price tags.

8
00:02:32.113 --> 00:02:52.742
Financial comparisons between the two companies carry important nuances. Anthropic reported higher second-quarter revenue and reached adjusted profitability, while OpenAI recorded a widening operating loss, though OpenAI notes that Anthropic counts sales made through cloud partners differently in its top-line figures.

9
00:02:53.092 --> 00:03:18.820
A draft of Anthropic's prospectus showed an eighty billion dollar first-half net loss, though seventy-two billion of that was a non-cash paper adjustment for stock grants, leaving an operating loss of eight point seven billion dollars. The documents also outline up to seven hundred fifty billion dollars in long-term compute commitments, with four hundred fifty billion in uncancelable obligations spread over seven to ten years.

10
00:03:19.170 --> 00:03:33.485
Meanwhile, investors debating Anthropic's debut have reportedly pushed for a valuation around one point five trillion dollars rather than a rumored two trillion, though neither figure has been confirmed by the company or its bankers.

11
00:03:35.884 --> 00:03:53.676
Examining the market structure behind this timing, reporting emphasizes that OpenAI's deadline is effectively shaped by its competitor's listing schedule. Closing before mid-November allows the company to secure terms before open public trading sets an explicit market price.

12
00:03:54.026 --> 00:04:09.791
While discussions suggest the round could price OpenAI above one point four trillion dollars, no formal papers have been signed. The broader concern among investors is that public equity markets may prove far more cautious than private venture funds.

13
00:04:10.141 --> 00:04:28.893
The outcome carries broader ripple effects: major technology partners like Amazon have recorded large pre-tax gains on paper from private A.I stakes, meaning that if Anthropic lists at the lower end of expectations, those accounting valuations could face downward pressure across the sector.

14
00:04:31.293 --> 00:04:48.402
As it prepares for the public markets, Anthropic is also pushing its software directly onto user devices. The company announced a preview of Claude Duet, a personal assistant built on Claude Opus five point five that is available through web and mobile applications.

15
00:04:48.752 --> 00:05:05.456
Unlike conventional chatbots that wait for a question, Duet is designed to work proactively. It delivers a morning task briefing, monitors permitted services in the background, and requests approval to handle assignments ranging from scheduling to building web applications.

16
00:05:05.806 --> 00:05:28.206
Through integrations, Anthropic says the tool can find missed messages, draft replies, publish job listings, handle payroll workflows, set up storefronts, and complete business purchases. This marks a strategic expansion from enterprise software toward consumer and professional desktops, competing directly with proactive tools from Meta and OpenAI.

17
00:05:28.556 --> 00:05:47.713
On privacy, the company says automated classifiers reject requests to share strict credentials like government identification or credit card numbers, while health and financial records require explicit user consent. For now, access is gated behind a waitlist restricted to paying subscribers.

18
00:05:50.112 --> 00:06:12.598
Turning from office tools to wet-lab science, Anthropic has released research on an experiment testing whether software could help design treatments for scorpion venom. An unreleased version of Claude generated over three hundred thousand candidates across sixty-four hours, using specialist tools to select designs that were then sent for physical synthesis.

19
00:06:12.948 --> 00:06:25.684
The project was carried out in partnership with Citrini Research, an investment research firm that spent roughly two hundred thousand dollars and whose participating staff had no formal background in drug development.

20
00:06:26.034 --> 00:06:46.962
Laboratory evaluations confirmed that seventeen of the two hundred eleven tested proteins successfully bound to targeted toxins. However, Citrini described the universal antivenom experiment as a failure because the molecules were never tested to see if they could actually neutralize venom in living cells or protect animals.

21
00:06:47.312 --> 00:07:08.474
The project illustrates clear development limits: while computer models generated candidate proteins quickly, the team spent about fifty thousand dollars on compute and nearly one hundred fifty thousand on physical lab work, waiting two months for testing. Proving that a molecule physically binds in a dish remains a very long way from producing a safe medicine.

22
00:07:10.874 --> 00:07:28.090
In clinical diagnosis, Microsoft researchers published a study in the journal Science demonstrating a multi-agent system called MAI-DxO, which acts like a panel of consulting doctors that question patients, order tests, and debate a final diagnosis.

23
00:07:28.440 --> 00:07:52.482
In simulations using complex case records from the New England Journal of Medicine, the orchestrated system reached the correct diagnosis eighty-three percent of the time, compared to thirty-six percent for individual generalist physicians using reference materials. It also outperformed human doctors on emergency department admission records, while estimating lower diagnostic costs.

24
00:07:52.832 --> 00:08:14.635
While Microsoft A.I leadership described the findings as a step toward medical superintelligence, the study carries important boundaries. The benchmark compared the software against solo primary care doctors rather than the teams of specialists who normally manage complex cases, and it evaluated older foundation models rather than the newest frontier systems.

25
00:08:14.985 --> 00:08:37.492
The study's own data also revealed that foundation model capability drove most of the accuracy, with the agent setup adding incremental gains that narrowed as models advanced. Microsoft now plans a prospective clinical trial in a Boston emergency department, comparing recommendations alongside attending physicians for consenting patients.

26
00:08:39.891 --> 00:08:56.211
International funding remains equally substantial. Bloomberg reports that Chinese artificial intelligence developer DeepSeek has secured about fifteen billion dollars, or roughly one hundred billion yuan, in one of the country's largest private funding rounds.

27
00:08:56.561 --> 00:09:14.118
Prospective backers reportedly include electric vehicle battery maker CATL and WeChat owner Tencent. The funding follows an earlier seven billion dollar round in June that valued DeepSeek at fifty billion dollars, as the company explores plans for a domestic listing next year.

28
00:09:14.468 --> 00:09:36.399
The reported capital push comes just after United States civilian agencies were directed to remove DeepSeek software over security concerns. DeepSeek did not comment on the financing, but the transaction underscores how domestic corporate leaders continue funding domestic computing infrastructure and open models regardless of Western restrictions.

29
00:09:38.798 --> 00:09:56.398
In hardware manufacturing, Intel is reportedly partnering with Tesla and SpaceX on a joint computing initiative called Terafab. The project is described as combining Tesla's workloads, SpaceX's infrastructure, and Intel's chip design and manufacturing operations.

30
00:09:56.748 --> 00:10:09.526
The tie-up introduces Intel into Elon Musk's broader aerospace and automotive ecosystem, offering visible demand to complement Intel's heavy capital spending on advanced manufacturing nodes.

31
00:10:09.877 --> 00:10:24.148
Industry observers caution, however, that the commercial reality of the project remains speculative until Intel discloses confirmed silicon production volumes or binding design wins rather than high-level partnership headlines.

32
00:10:26.548 --> 00:10:37.342
Finally, developer Nate Herk demonstrated an experimental workflow using Claude Code that sets up an adversarial council of sub-agents to critique business decisions.

33
00:10:37.692 --> 00:10:57.788
Dubbed the Roast Room, the system assigns specific evaluation frameworks to simulated advisers modeled on public figures. For instance, a simulated Warren Buffett evaluates cash risk and long-term value, while other seats focus on pricing, willingness to pay, and whether operations can survive without founder intervention.

34
00:10:58.138 --> 00:11:18.938
The setup organizes persona debating styles, principles, and knowledge into structured files, allowing users to build custom council members. While the code is shared through the creator's community platform, simulating executive perspectives remains a creative brainstorming exercise rather than validated strategic advice.

35
00:11:19.288 --> 00:11:35.736
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.