WEBVTT

1
00:00:10.000 --> 00:00:24.869
Welcome to the Global Advisors A.I Voices podcast. I'm Maya.
Today is Sunday, the twenty-second of November 2026.
This is your morning catch-up on A.I. Please like and follow the show for your daily A.I update.

2
00:00:26.918 --> 00:00:51.238
In a recent analysis, the Nate Herk A.I Automation channel states that spending on computer equipment and software accounted for about four percent of the United States economy in the first half of last year, yet represented approximately ninety-two percent of the economy's growth. This disproportionate impact highlights how heavily the current expansion relies on the A.I infrastructure build-out.

3
00:00:51.588 --> 00:01:12.559
The channel also explains that much of this spending moves in a loop known as circular financing, involving investments from chipmakers like Nvidia to companies like OpenAI and Anthropic, which then purchase chips. This dynamic creates a concentrated source of demand that can amplify both growth and potential instability.

4
00:01:12.909 --> 00:01:32.365
PYMNTS reports that real G D P grew at an annual rate of three point four percent in the third quarter. The Bureau of Economic Analysis attributed this acceleration to resilient consumer spending and business investment in artificial intelligence, particularly in information processing equipment.

5
00:01:34.764 --> 00:01:55.799
In a recent video, Nate Herk from A.I Automation highlighted early access reports regarding Gemini 4 Neon's coding capabilities. He noted that a user named Chubby reported the model built seventeen interactive demos, including a destructible voxel arena, demonstrating its ability to handle complex creative structures.

6
00:01:56.149 --> 00:02:12.981
Herk also shared Chubby's experience with debugging, where the model successfully repaired broken programs. According to the report, Neon raised the test pass rates for several applications to one hundred percent, effectively resolving issues that had previously caused failures.

7
00:02:13.331 --> 00:02:28.819
These anecdotes suggest that Gemini 4 Neon is particularly effective at code repair and debugging tasks. While these individual successes are promising, they remain user-reported observations rather than comprehensive benchmark results.

8
00:02:31.218 --> 00:02:56.605
Baseten has released bison-1 and groundhog-1, open-weight decision models built on Google's Gemma 4. Unlike standard language models, these are designed to return calibrated probabilities for yes/no, choice, and score questions rather than generating text. Bison-1 scored seventy-six on the Artificial Analysis Decision Index v2.1, ranking first among the models evaluated.

9
00:02:56.955 --> 00:03:14.021
Both models support a context window of one hundred thirty-one thousand seven hundred twenty tokens and accept images. Baseten says this represents a significant increase over the context limits of its previous models, allowing for more complex decision-making inputs.

10
00:03:14.371 --> 00:03:33.955
Bison-1 is priced at ten cents per million tokens, while Groundhog-1 is priced lower. Baseten attributes calibration gains to soft-label targets and trained the models using B200 GPUs. Groundhog-1 has not yet received an independent Artificial Analysis Decision Index score.

11
00:03:36.355 --> 00:03:52.141
Signal has announced it is limiting users' ability to take screenshots within its secure messaging app on both Android and iOS. The move is framed as a defense against what the company describes as an existential threat from A.I surveillance.

12
00:03:52.491 --> 00:04:11.926
The announcement was posted on November twenty-first, 2026. By restricting screenshots, Signal aims to prevent the easy capture and subsequent A.I-driven analysis of private conversations, reinforcing its commitment to privacy in an era of increasingly sophisticated surveillance tools.

13
00:04:14.325 --> 00:04:36.085
OpenAI has launched GPT-Live-1.1, a full-duplex voice model for multilingual conversations available via its API. The company states the new version features more natural and consistent accents across languages, more accurate speech transcription, and better handling of precise details like letters and numbers.

14
00:04:36.435 --> 00:04:54.099
A key addition is standby mode, which bills idle WebRTC time at two cents per minute, compared to five cents per minute for active sessions. Standby mode pauses inference during silence while keeping the connection open, with a configurable idle timeout starting at five seconds.

15
00:04:54.449 --> 00:05:16.806
OpenAI claims performance improvements over GPT-Live-1 but has not published benchmark scores for the new version. GPT-Live-1.1 is available on Vercel A.I Gateway and maintains compatibility with existing integrations, offering a more cost-effective option for voice agents that spend significant time in idle states.

16
00:05:19.206 --> 00:05:38.683
Cloudflare C E O Matthew Prince warned that A.I-powered vulnerability discovery could cause a two-year stretch of outages. A.I lowers barriers for both defenders and attackers, allowing flaws to be found and weaponized faster than before, forcing companies to patch systems at an unprecedented pace.

17
00:05:39.033 --> 00:05:57.828
Prince stated that Cloudflare knows of numerous serious vulnerabilities beyond recent PHP issues, suggesting the worst may still be ahead. Recent technical outages have already affected major services including Outlook, Apple, United States banks, United Airlines, and the NSA.

18
00:05:58.178 --> 00:06:08.908
Diego Almada Lopez argues that this warning may signal a shift in how the cybersecurity sector is valued, with stakeholders reassessing risk management strategies.

19
00:06:11.307 --> 00:06:34.198
The European Central Bank's November financial stability review argues that advanced A.I models are increasing the scale and speed of cyberattacks, potentially favoring attackers in the near term. The report notes that low-cost A.I agents make sophisticated attacks feasible for less specialized actors, lowering the barrier to entry for malicious activity.

20
00:06:34.548 --> 00:06:56.543
According to the ECB, defenders are structurally disadvantaged because they face higher costs for deploying frontier models and are bound by slower, regulation-heavy decision-making processes. While A.I tools can help defenders detect vulnerabilities, the report suggests that the speed advantage currently lies with those who can act without such constraints.

21
00:06:56.893 --> 00:07:19.143
The ECB Financial Stability Review forecasts that even if systems become more secure, cyber risks will persist. It outlines scenarios ranging from continuous vulnerability discovery to a potential convergence on safer systems through 'security by design,' but emphasizes that the immediate threat landscape remains skewed toward attackers.

22
00:07:21.543 --> 00:07:47.185
Nate Herk from the A.I Automation channel analyzes Anthropic's recent IPO, highlighting a circular financing structure involving major tech players. He points out that Amazon and Google own significant stakes in Anthropic while simultaneously selling it computing power and buying its services. Additionally, Broadcom has agreed to lend the company up to forty-two billion dollars to help pay for chip leases.

23
00:07:47.535 --> 00:08:08.485
Herk notes that seven chip and cloud companies, including Nvidia, purchased twenty-six billion dollars of Anthropic stock during the IPO. This interlocking investment pattern means that the same entities are funding the lab, supplying its infrastructure, and buying its output, creating a concentrated web of financial dependency.

24
00:08:08.835 --> 00:08:24.664
The channel argues that this loop relies on continued explosive demand for A.I. Herk warns that if demand slows, the financial benefits could reverse for all participants simultaneously, exposing the fragility of this interconnected funding model.

25
00:08:27.063 --> 00:08:46.668
Nate Herk reports that Gemini 4 Neon costs 25 percent less than Anthropic's Opus 5.5 standard rates. However, in a watch configurator test, Herk found that Neon cost 1.86 dollars but produced a lower-quality result compared to Opus at 2.74. dollars

26
00:08:47.018 --> 00:09:03.744
The source video states that Neon scored slightly above Opus 5.5 on the Artificial Analysis leaderboard. The source authors note that Neon would need to use significantly fewer tokens than competitors to erase its price premium in some comparisons.

27
00:09:04.094 --> 00:09:19.198
Nate Herk's testing highlights that lower per-token costs do not guarantee lower per-task costs. Developers must evaluate specific workflows, as Neon's efficiency gains may not offset higher token usage in complex reasoning tasks.

28
00:09:21.597 --> 00:09:42.717
Thomas Wolf, speaking on The MAD Podcast, argues that open source A.I provides a critical accessible path for biotech startups. He notes that commercial A.I access for life science companies is currently very limited due to strict guardrails and concerns regarding biohacking, which has led some companies to switch to other options.

29
00:09:43.067 --> 00:10:02.651
Wolf believes that open source models allow new companies to find their own way to start building their own A.I independently. He suggests this approach fosters a diversity of new ideas, bypassing the restrictions imposed by major commercial providers who may limit access to sensitive biological data.

30
00:10:03.001 --> 00:10:25.550
This shift is enabling minimalist biotech companies with very few employees, sometimes as few as one, to become more feasible. Nature reports that the rise of A.I tools is allowing these small teams to achieve what previously required large institutional resources, further validating Wolf's point about the democratizing potential of open source.

31
00:10:27.949 --> 00:10:46.125
Shortical has launched a face licensing program for microdrama actors, debuting with the A.I-generated series 'Inevitable' starring Israeli actor Aki Avni. The startup is transitioning to A.I content to lower production overhead while partnering with Payouts.com for creator payments.

32
00:10:46.475 --> 00:11:05.953
Industry consultant Jennifer Cooper warns that A.I has devastated the microdrama industry, noting that compensation structures remain unclear. She cites reports of payments ranging from fifteen to seven hundred dollars in China, highlighting the potential for significant undervaluation of actors' likenesses.

33
00:11:06.303 --> 00:11:24.607
Questions persist regarding consent and royalty-based compensation. Company representative Shimoni stated that payments are handled on a case-by-case basis, leaving uncertainty about how actors will be fairly compensated as this trend potentially expands from China to Hollywood.

34
00:11:24.957 --> 00:11:41.405
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.

