October 6, 2026
2026-10-06
https://www.youtube.com/watch?v=ig2isKaLljg

[0:10] We go live at ten.
[0:24] Every morning, the future moves fast.
[0:29] The mosque, money, the power, and the people building what comes next.
[1:07] From Austin, Texas to the world.
[1:14] AI News, Real Context.
[1:18] This is the Austin Wilson Show.
[1:20] I'm Austin Wilson, your human in the loop.
[1:30] Good morning, and welcome to the Austin Wilson show.
[1:34] It's 10/06/2026.
[1:38] Welcome.
[1:39] We are breaking the AI headlines that matter to you today.
[1:43] I'm Austin Wilson coming to you live from the AI capital of the world, Austin, Texas.
[1:48] And today, we've got an exciting show for you.
[1:51] We're covering AI labs testifying under oath in New York City.
[1:55] ChatGPT text gets a watermark.
[1:59] AI now writes prescriptions in Utah.
[2:02] XRock engineers sue over Nvidia's $20,000,000,000 deal.
[2:07] Etch reportedly fields bids of up to $50,000,000,000, and US startup takes on China's open models.
[2:16] Let's get into it.
[2:18] First up, we are talking about OpenAI, Anthropic, Google, and Meta sent executives to testify under oath before the New York City Council, and only after the council threatened subpoenas.
[2:31] We saw subpoenas recently out of California last week.
[2:35] Wonder if it'll lead to a similar outcome.
[2:37] All 51 council members sat as a committee of the whole on Monday, asked the chance of an AI catastrophe, OpenAI's.
[2:46] Morgan Dwyer said, I don't know.
[2:49] I also don't think it matters whether it's 1% or 10% or a 20% chance adding that no level is acceptable.
[2:56] Speaker Julie Menon called that flipping.
[3:00] Anthropic's Logan Graham and Meta's Shane Cahill gave no number.
[3:05] Google's Alex Friend said no rigorous method exists yet, and former, uh, Google DeepMind researcher Alex Turner put the odds of an eventual AI takeover at roughly one in three.
[3:20] Wow.
[3:20] These are, uh, interesting numbers.
[3:22] Uh, city council got the labs on the record under oath, and none of them would put a specific number on a catastrophe or say who pays if an AI goes rogue.
[3:35] The company say they test their models for risks from cyber attacks to loss of control.
[3:41] On liability, OpenAI, Anthropic, and Medic gave no direct answer.
[3:46] Only Google was clear, and they said if it's illegal without AI, it's illegal with AI.
[3:52] Menon demanded written follow ups.
[3:55] The bills on the table are city bills, a whistleblower reward program, a right to sue over harm from AI agents, and independent testing.
[4:05] The models are built far outside of New York City.
[4:09] Will their jurisdiction have an effect here?
[4:11] This is interesting to see as we saw Gavin Newsom signed some bills last week around AI and AI compliance and and how these frontier companies must act in the state of California.
[4:28] We also saw a subpoena issued by the state of California, which we suppose might lead to a similar kind of counsel here that we're seeing with New York City, and getting some interesting takes.
[4:42] The the what does this matter though?
[4:45] Like, don't understand why they're pressing them on what are the odds that the world is going to end.
[4:50] The fact of the matter is this technology is here.
[4:52] It is fantastic.
[4:54] It is going to change everybody's lives, and net net on net, it's gonna change lives for the better.
[5:01] So this is a a benefit to society.
[5:03] This is something that's excellent.
[5:06] Do we know exactly where it's gonna go?
[5:09] No. We we don't.
[5:10] But that's the important thing is we start to see safety and guardrails built around the technology, which helps keep it in check as we do develop it.
[5:19] Nobody knew where the Internet was going thirty years ago, and yet we created an open environment where people could go in, use the Internet, use the technology, and uncover vast opportunities for individuals, for companies, for countries.
[5:36] And it...
[5:37] The Internet has changed our lives night and day compared to thirty days...
[5:42] Thirty years ago.
[5:43] And it's all to the benefit of society as a whole, and we expect to see the same thing with AI here.
[5:49] And so when you ask these executives at frontier companies what their number is on the chance that AI is going to end the world, it's...
[5:58] We...
[5:58] What are we talking about?
[5:59] We're focusing on the wrong things here.
[6:02] We should be focusing on how to, um, skill up everybody that we can, and get people using AI and understanding how to use AI and how AI can change their lives.
[6:14] So sad to see this coming out of New York City and the questions here.
[6:18] And I think moving forward, you know, I think about if the state of Texas did something like this.
[6:25] It's like, what does this matter?
[6:26] They don't have any jurisdiction around the development of of this software, of this technology.
[6:33] And so to suppose that maybe New York would, you know, strap some sort of restriction on the the OpenAI or or or Anthropic or or or Meta or Google Labs is a little bit absurd.
[6:46] And if anything else, they were...
[6:48] They would limit the the possibilities and capabilities of their own people.
[6:54] So I hope that that doesn't occur.
[6:56] Moving on to our next story.
[6:59] OpenAI will start hiding in a visible watermark in the text ChatGPT and Codex, uh, that that is written for users in the European Union.
[7:12] OpenAI said Monday that over the coming weeks, eligible ChatGPT and Codex output in The EU will carry text grain, a statistical signal hidden in the model's word choices.
[7:25] Readers can't see it.
[7:26] A detector can.
[7:28] API customers worldwide can switch it on, but it's off by default.
[7:32] OpenAI plans to open source the technology, and at first, only approved researchers and organizations get the detector.
[7:43] This is a curious story.
[7:45] Teachers, employers, platforms want a way to tell AI writing from human writing.
[7:51] ChatGPT now joins Anthropic's Claude and Google's Gemini in marking its own text.
[7:56] The claim, OpenAI says this means that EU AI acts transparency rules for AI generated content.
[8:04] The evidence, OpenAI's own tests at a 1% false positive rate, the detector found the mark in about 95% of 400 token passages, about 80% of 200 token passages.
[8:15] The catch, swapping 25% of the words, only 25% of the words for synonyms drop detection to 17%, and OpenAI says a missing watermark quote unquote does not prove human authorship.
[8:28] Let's get into it.
[8:30] To me, this makes zero sense at all.
[8:33] I I understand, you know, protecting against, know, false flags or or AI, you know, misleading people or or anything like that, any dangers around this.
[8:46] But at this point in time, AI is extremely detectable.
[8:50] AI writing is very, very detectable by humans, um, and forcing the the labs to put a watermark on this.
[8:57] I think this is just an example of governments may...
[9:00] Moving way slower than, uh, than technology and really than than the market.
[9:06] Um, this might have been impactful six months ago or or something like that, but there are already solutions in, uh, the market available.
[9:15] Pangram, uh, users can use to identify AI writing.
[9:20] I think I think this becomes more important when you're talking about deep fakes, fake videos, fake voice, fake images that look indistinguishable to a real life human, or sound like a real life human, and that becomes dangerous.
[9:39] And having watermarks on that, I think, become really important.
[9:43] And I believe that those watermarks are already in effect or, you know, already in place.
[9:50] Even still on that front, we don't need the government to to tell us to watermark those things because there are already technologies that can identify AI generated image, AI generated video, and and voice because of the pattern that these technologies use.
[10:08] We're able to detect that.
[10:09] So, you know, while I appreciate the safety effort coming out of the EU, and again, I appreciate how slow OpenAI is to the, you know, the last ones to adopt any safety measures, and yet they seem to be the ones proliferating a lot of the safety issues.
[10:29] But to me, it doesn't really have much of an impact, and I can guarantee you that we'll continue to see, you know, headlines around safety efforts and safety mechanisms that really actually don't mean anything, unfortunately.
[10:48] This is a fun one.
[10:49] Utah is letting a, uh, startups AI write acne prescriptions on its own.
[10:57] Uh, and after it writes prescriptions for its first 100 patients, no doctor needs to be involved before the prescription goes out, which is kinda crazy.
[11:06] NOLA Health said Monday, its NOLA derm app can issue topical acne prescriptions to Utah adults under an agreement with the state's office of artificial intelligence policy at a pilot price of $4.99 a month.
[11:22] Wow.
[11:23] Two physicians approve every prescription for at least the first 100 patients, then the AI prescribes directly with weekly doctor review up to 500 patients.
[11:34] After that, doctors audit at least 10% each month.
[11:39] Wow.
[11:40] No pills, no extra review, no severe acne.
[11:46] Founder Luis Winis, until now, says, until now, a licensed clinician signed every single prescription before it went out.
[11:56] Our AI can now make that prescribing decision itself.
[12:01] Wow.
[12:02] This is fascinating.
[12:03] Uh, currently, this is this is a template.
[12:06] Right?
[12:06] NOLA's medical advisor already names the flu and uncomplicated urinary tract infections as possible next steps.
[12:16] NOLA says the AI chooses among preapproved treatment protocols rather than making free form decisions.
[12:24] The stage doctor oversight is spelled out in NOLA's announcement and SiliconANGLE's report.
[12:31] The waiver runs for one year.
[12:33] Okay.
[12:34] So today, it's acne cream.
[12:35] The real question is who signs off when the list of conditions grows?
[12:41] Let's get into it.
[12:42] This is very interesting.
[12:44] I love to see AI and this technology being used in a new use case, a new field.
[12:51] We haven't seen a lot used in in direct clinical practice yet, so this is really interesting and exciting.
[13:02] And a really great use case where I imagine the majority of these prescriptions are refills and folks just coming in to say, hey hey, doc, you know, I I need a refill on my acne prescription.
[13:17] Let's get that filled.
[13:18] And and it takes the doctor's time and and effort to review these, whereas it's pretty standard case by case.
[13:29] Really interesting to see the escalation, you know, the doctors involved in the first 100 patients, and then it scales to, you know, five hundred and then checking, you know, a certain percentage beyond that.
[13:44] So it's interesting to see the scale where we have human in the loop and then gradually move the doctor out, uh, where AI is predominantly, uh, taking over and making the decisions.
[13:56] And and really interesting to see how efficient this will make doctors and hopefully, uh, free these doctors up to spend more time on more serious conditions and provide a higher level of of care to, uh, other patients.
[14:12] Um, I think probably the concern here, maybe the the obvious, uh, thing is, um, does this increase the rate of prescriptions being written?
[14:22] Um, if you adopt this to to additional modalities from a, you know, a prescriber's perspective, does that get really dangerous when we've had issues with prescription drugs in this country historically.
[14:40] Now do we have AI just, uh, kind of freewheeling and writing prescriptions that it shouldn't be?
[14:48] It seems like they've got the guardrails in place.
[14:50] They've got the oversight, and we'll be reviewing these cases.
[14:54] But you've got to imagine that this incentivizes doctors that, you know, maybe they do have a financial incentive for writing some of these prescriptions, and now they just went from having to manually review and write, uh, you know, a hundred, five hundred, a thousand prescriptions.
[15:13] And now they can have AI do that and, you know, presumably benefit without having to, you know, do a linear amount of work.
[15:26] So interesting to see this.
[15:30] Interesting to see the use case.
[15:32] I think this all in all will have a net positive impact on the medical community, and we hope to see more advancements like this.
[15:41] And and shout out to Utah.
[15:43] I mean, what a really cool cool that they have an AI policy department.
[15:51] Uh, Congrats to Utah there, and congrats to the team at NOLA.
[15:56] Uh, as we move on to the next story, two former Grok, g r o q, engineers are suing Grok's board saying Nvidia's $20,000,000,000 deal for its tech and talent shortchanged ordinary shareholders.
[16:16] Benjamin Sarah Bryn and Joshua Rubin filed suit Monday in Delaware's Court of Chancery against Grok's board naming founder Jonathan Ross and president new money Sunny Madra.
[16:28] In December 2025, Nvidia paid about $17,000,000,000 in licensing fees, plus roughly $3,000,000,000 in stock bonuses to bring about 200 Grok engineers over.
[16:43] It's about $20,000,000,000 in all in a deal called a a non exclusive.
[16:49] The suit says the board skipped the stockholder vote, uh, that Delaware law requires and left Grok a hollowed out shell.
[17:01] What happened here?
[17:03] This, uh, is a quote, unquote license and hire move that big tech, um, uses to buy AI startups without technically buying them.
[17:14] If a court says it needs a shareholder vote, the playbook gets much much more difficult.
[17:19] Right?
[17:20] So the justice department is already looking at this deal.
[17:23] The New York Times reported this back in September.
[17:26] The plaintiffs say board conflicts left common stockholders billions of dollars worse off.
[17:33] The complaint, as reported by the Financial Times and law.com's Delaware Business Court, these are allegations.
[17:40] Grok and Nvidia hadn't responded in the reports, and both plaintiffs left Grok before the deal, uh, but they kept their stock.
[17:50] Let's get into this one.
[17:52] Okay.
[17:52] This is interesting.
[17:53] Um, sorry to the fellas that quit Grok before the Aqua hire occurred, but it looks like they might be getting the short end of the stick here, and maybe a good reason, uh, to to be loyal and stick around.
[18:07] I don't know.
[18:09] You know, unfortunately, or maybe fortunately, we see...
[18:12] Have seen these more recently.
[18:14] This happened at Meta.
[18:16] This would be called an aqua hire where, uh, a larger business goes in and instead of buying the...
[18:24] A company and, you know, doing the general and and typical mergers and acquisitions process, they'll just say, hey, let let me hire the 200 people that matter at this company.
[18:35] Bring them over.
[18:36] And in this case, at Grok, they, uh, licensed the Grok technology for $20,000,000,000.
[18:45] It's exclusive license, and I believe, I'm sure has a a really, uh, long dated expiration on it.
[18:51] So that essentially, in in effect, they own it, but they don't have to buy out the common shareholders.
[18:58] And this is, you know, kind of a bummer, but it's kind of the way the world works as well too.
[19:03] It's like, you know, first money in, first money out.
[19:05] So the the investors, the the founders of Grok, they have preferred shares, and I'm sure they got paid out on this deal.
[19:15] And and I believe it was a a great deal for these folks, including, you know, of the All In podcast, uh, and others.
[19:24] But when you talk about common employees that held some shares, of course, I don't know the details.
[19:29] You know, how much how much stock did these fellas own?
[19:33] Um, how long were they at Grok?
[19:35] Were they there for three months?
[19:36] And, uh, and then, you know, we're hoping that they're deal vested.
[19:40] I don't know.
[19:41] This is, uh, unrecorded yet, but, you know, it seems like maybe they missed the boat and they see a $20,000,000,000 payday and they say, hey, let's see if we can go get a piece of this.
[19:56] Interesting to note, um, this is in the state of Delaware where, um, in past years, there's been conflict with Elon Musk and Tesla and Tesla shareholders and Elon Musk's pay package.
[20:09] And, you know, obviously, when you get really big numbers with lots of zeros and a handful of commas, you you get a little bit opportunistic, I think.
[20:21] And so can't blame these guys for trying, I guess, and and you gotta expect this sort of thing in, you know, corporate mergers and acquisitions or acqui hires, and that that there is gonna be some litigation here.
[20:34] So we'll see where this story goes.
[20:36] Wish best of luck to those guys.
[20:38] Uh, wish a little bit more luck to Grok and the team at Nvidia because they are the ones pushing technology forward instead of, uh, trying to distract these folks from, uh, getting their work done.
[20:51] Uh, here we go.
[20:53] This one's great.
[20:54] Chip startup etched is reportedly fielding offers that value it at 40 to $50,000,000,000.
[21:02] Uh, that's about double the $21,000,000,000 that it was worth just one month ago.
[21:08] TechCrunch reported Monday, citing sources that established investors offers value etched at about $40,000,000,000, and lesser known, uh, offers have come in valuing it about $50,000,000,000.
[21:24] Etched was valued at 10 after a $300,000,000 round led by Sequoia, and at $21,000,000,000 in September when Jane Street led $700,000,000.
[21:41] My goodness.
[21:42] This is great to see.
[21:43] Nearly five fold in three months shows how hard money is chasing alternatives to Nvidia for running AI.
[21:53] Uh, or is it the fact that these folks were on with Patrick O'Shaughnessy on invest like the best, and and and now they're just getting a a skyrocket, just going vertical from that audience.
[22:07] Who knows?
[22:08] Etch says its chips process tokens faster and cheaper than Nvidia.
[22:13] Really cool story, by the way.
[22:15] Look up the Etch story.
[22:16] Go listen to that podcast on invest like the best with, uh, Patrick.
[22:20] Uh, really cool to see what these guys have done.
[22:24] Young guys, super smart, and have gone to hustle.
[22:28] The report said in July that it had a $11,000,000,000 in customer orders at etched, and Jane Street is both an investor and a customer.
[22:40] Also really cool what Jane Street is doing.
[22:43] No round has closed, and the report doesn't say who's bidding or whether the money goes to the company directly or what that process is.
[22:53] This is really interesting, um, just on a number of different planes.
[23:01] Super exciting, and huge congrats to the team at Edge for what they've accomplished.
[23:08] And just to see the answer in the marketplace, what their technology is enabling and what they focused on and have been working on for several years now is is really paying off and just, you know, really a a very cool outcome.
[23:26] And to see this happen this fast to go from, you know, $1,000,000,000 in orders at the beginning of the summer to 10,000,000,000, 20,000,000,000, 40 and $50,000,000,000 valuations is is phenomenal.
[23:39] So really exciting.
[23:40] Why does this matter?
[23:41] Why is this important?
[23:43] To put it simply, etched has a unique process of building chips that our AI runs on, that is more efficient and faster than traditional Nvidia chips.
[23:57] Nvidia, as we know, is the they're they're the they're the big boss in the space.
[24:06] Right?
[24:07] Phenomenal company, have done a wonderful job, have been around for thirty plus years, and have led the AI wave and and opened up a lot of capabilities through their different chip platforms and and and software models that enable, um, AI companies to build these advanced models.
[24:28] Because of their success, um, we we see competitors arising to try to figure out how to do a better job than Nvidia and seeing the opportunity therein and case in point, etched.
[24:42] So these guys, um, obviously, their technology is, uh, something that is really impactful and doing some amazing things for its customers.
[24:53] Jane Street is a a very interesting company.
[25:00] Very, very smart people work there, and they are using etched etched...
[25:07] Wow.
[25:07] That's difficult to say.
[25:08] They're using the chips that etched created to build their own AI on top of...
[25:16] To use that in trading in markets as they are a quant firm, and this is a big piece of their business.
[25:24] So just really cool to see this story, really cool to see, uh, the success here, and awesome to see these guys pushing the envelope forward.
[25:34] The question is, you know, to piggyback on our our, uh, previous story about Grok and that acquisition there, obviously, Nvidia is willing to spend $20,000,000,000 on a chip company.
[25:46] Are they looking at a $50,000,000,000 acquisition of etched?
[25:53] Would that be a natural acquisition play for them?
[25:57] Would Etched sell to them?
[26:00] Or do they wanna keep all the chips on their end of the table and keep stacking and see where they go from there?
[26:07] $50,000,000,000 is a pretty phenomenal valuation.
[26:11] Of course, we did see the Cursor acquisition for $60,000,000,000 earlier this year.
[26:16] So there's...
[26:18] You know, that's not off the table for Nvidia, I would say.
[26:22] But it's interesting to see where it goes from here.
[26:25] We'll continue to track this story and report on, uh, any interesting outcomes as they develop.
[26:34] Reflection AI unveiled unveiled Beam.
[26:38] Wow.
[26:39] Okay.
[26:39] A 501,000,000,000 parameter open model that it says matches top Chinese open models on a fraction of the hardware.
[26:49] Beam is a mixture of experts model, uh, 501,000,000,000 parameters in total, 23,000,000,000 active per token.
[26:59] Wow.
[26:59] Trained on 23,800,000,000,000 tokens.
[27:02] Reflection says it beats GLM 5.2 on some tasks using a third to a quarter of the hardware and approaches Quinn 3.8 max, while still trailing frontier models like Anthropic, Claude, Fable 5.1.
[27:19] Early access opened Monday.
[27:22] The waits come later in October under the Apache two point o license after red teaming.
[27:29] Reflection was recently valued at $25,000,000,000.
[27:34] This is phenomenal.
[27:35] What an exciting story to talk about here.
[27:38] Open Weight AI has been China's game.
[27:41] This is the strongest US bid to change that if the weights actually ship.
[27:46] Uh, reflection says Beam is the first open model from a US startup to match the best Chinese models.
[27:53] Uh, reflection's own benchmarks, independent tests, wait for, uh, the actual weights to come out, and then we'll be able to test this with third part...
[28:03] Third party testing and see where that lands.
[28:05] Uh, you can't download it yet.
[28:07] It looks like the weights are, uh, planned for later this month.
[28:12] And let's get into this topic.
[28:17] This is a little bit more technical, so we'll try to explain it.
[28:20] But net net, the really exciting thing here is open source.
[28:25] And why does open source matter?
[28:27] And why does US open source matter?
[28:30] There's a number of reasons here.
[28:32] We we want technology.
[28:34] We want this intelligence to be easily distributed and open and available to everybody.
[28:40] And when you have constraints around, you know, whether it's gating who can access your your models, protecting against foreign access, or preventing people based on financial ability, whatever it may be.
[29:02] We want to have open source so that everybody has access to the same technology.
[29:08] It offers a level playing field.
[29:10] Nobody gets intelligence greater than somebody else, and that's the idea here.
[29:16] And so when we see a US open source model, uh, come out that has, uh, near frontier level intelligence, this is really important.
[29:25] Frontier is...
[29:26] Think of Anthropic and OpenAI.
[29:30] They are frontier.
[29:31] They are leading the, uh, the the pack with the most intelligent, most advanced technology.
[29:38] But you move down just a little bit and you have some really phenomenal open source technologies out of China, uh, that are extremely capable, extremely low cost.
[29:50] The problem is they're based out of China, our adversary, and being able to ensure that those models remain open, remain accessible to The US.
[30:03] China kinda has a a history of offering open source solutions, gaining wide adoption, and then, um, making sure people are locked into those solutions, and then charging them for it, and it's hard to get off.
[30:20] And so when we see, uh, US model companies release an open model solution that is of the same technology, when we reference, uh, GLM or Quinn.
[30:32] Those are Chinese model companies.
[30:34] And so, um, when we see a US model, uh, come out that matches the advanced, uh, capabilities of Chinese models, that's really good.
[30:44] Get into a couple of the the numbers, I think, is interesting.
[30:48] Um, it's a mixture of experts model.
[30:54] What does that mean?
[30:54] It means that you've got a full model set.
[30:58] So, uh, the the the the model, the technology, the intelligence, think of it as a whole brain like this.
[31:05] The problem is, um, what you would do in this case, or in in most case with Llms is you're gonna ask it a question, right?
[31:16] When was Abraham Lincoln president?
[31:18] And it's gonna search all of that brain for that information.
[31:26] It it it's going to compute the whole...
[31:28] All the parameters within that brain to understand Abraham, like, a really simple question.
[31:35] Uh, Mixture of Experts actually decides, hey, let me split this up into sections, and then I'll only use the section that really matters for this specific question, instead of running compute across the whole thing.
[31:48] So it's much more efficient.
[31:50] Uh, and then to be able to run that on, I think it was 23,000,000,000 parameters.
[31:56] It sounds like a very large number, but, um, that's actually extremely small for, uh, these LLMs such that, um, an open source model like this, you could run on your own MacBook potentially.
[32:10] Very slowly, it wouldn't be, uh, you know, instant answers that you would get, but you could run it locally, which is really exciting for a bunch of different use cases.
[32:19] And and overall, you know, congratulations to, uh, the the the team at Reflection and and seeing this open source model come out, uh, is just an encouragement, I think, the rest of the open source community, to The US open source community, uh, to the AI community as a whole.
[32:39] And it's a big impact at some point in time.
[32:42] Hopefully, anybody will be able to go and download this for free and run it on their own computer and have technology and this intelligence at their fingertips, which is really exciting.
[32:55] Well, thank you guys so much for tuning in today.
[32:59] We had an amazing show.
[33:00] We covered a lot of really impactful topics, think, from what's going on with the New York City Council to Grok acquisition, etched success, um, and much more.
[33:13] We are continually researching and targeting the most important AI news that impacts you, and we wanna get it to you in a clear and understandable fashion.
[33:24] If you'd like more detail and more information, please go to taws.live.
[33:29] That's taws.live, the austin wilson show dot live, and sign up for our newsletter, The Loop, to make sure that you stay in the loop.
[33:42] Just remember, it's AI news, real context, and I'm your human in the loop.
[33:46] Austin Wilson.
[33:47] We'll see you tomorrow at 6AM central time.
[33:52] Let's go.