October 5, 2026
2026-10-05
https://www.youtube.com/watch?v=7DEo2BczGPo

[0:09] We go live in ten.
[0:23] Every morning,
[0:24] the future moves faster.
[0:28] The models,
[0:29] the money,
[0:31] the power,
[0:32] and the people building what comes next.
[1:07] From Austin, Texas
[1:09] to the world.
[1:14] AI news,
[1:15] real context.
[1:17] This is the Austin Wilson show. I'm Austin Wilson,
[1:21] your human in Luke.
[1:29] Good
[1:31] morning, and welcome to the Austin Wilson Show. My name is Austin Wilson. We will be breaking down the AI headlines that matter to you.
[1:41] Today, we're coming to you live from the super intelligence capital of the world, Austin, Texas. Today, Trump puts his spy chief over AI.
[1:51] Altman accepts some bad things, he says. Free Gemini users lose pro and flash models.
[1:58] Meta's muse reportedly
[1:59] profiles
[2:00] everyone you know, and then we've got a report Schneider nears a 20 plus billion dollar deal to acquire PTC.
[2:09] Let's get into it.
[2:12] First up,
[2:13] president Trump has named his top spy, director of national intelligence,
[2:18] Jay Clayton,
[2:20] to lead a new federal AI task force that he's calling the
[2:24] super intelligence force.
[2:26] Trump announced it Sunday on Truth Social,
[2:29] FTC chair Andrew Ferguson, Pentagon technology,
[2:33] and chief Emile Michael, and OPM director Scott Cooper
[2:38] serve alongside Clayton. Trump says the group will ensure that America continues to lead the world in superintelligence,
[2:46] and Clayton told the Wall Street Journal a report on AI's
[2:50] risks and opportunities
[2:51] is due
[2:52] within one hundred and twenty days.
[2:56] Now, America's AI policy is run by an intelligence official,
[3:02] which treats AI first as a national security issue. Clayton has already waived off calls to slow AI down. Here we go.
[3:11] Trump says the force will coordinate Washington's dealings with
[3:15] consumers,
[3:16] religious groups, infrastructure providers, and AI companies.
[3:20] Asked about slowing Frontier AI,
[3:23] Clayton said, I don't think any American should think that that's a good strategy.
[3:29] It's a coordinating task force with no regulatory
[3:32] power on top of a voluntary accord that companies wrote themselves.
[3:37] Now as we take a look at this latest advancement,
[3:41] we kinda have
[3:44] both sides of the force here.
[3:46] One
[3:47] side is pedal to the metal. Let's get things done.
[3:51] The other side is maybe more cautious, maybe
[3:55] doesn't know what they want out of AI or out of SI,
[3:58] super intelligence, as we look at it. I think the important thing here is to to see that we do have
[4:07] governments taking an interest in and ensuring, especially from this stance to ensure that we are moving the technology forward with... Which I think is really important that we are moving the technology forward. It's interesting that Trump put that line in there about religions
[4:23] as we've seen some
[4:26] AI folks talk about consciousness
[4:29] and
[4:30] and and AI having souls, which is...
[4:33] Seems absurd,
[4:36] but good to see that, hopefully, maybe we're having some logical people curtail some of these,
[4:45] I guess, less than normal ideas. Right?
[4:49] On top of that, interesting
[4:52] to have
[4:53] the director of national intelligence
[4:55] involved in this. I don't know how that plays into it. I don't know why
[5:00] other than maybe just because he already has intelligence in his title, and so it's just natural that he should lead up the super intelligence force.
[5:10] We'll see where that goes.
[5:12] It is... You know, you can't help but be a little bit disconcerted by the fact that
[5:17] the top spy is now leading the
[5:20] most powerful technology that we've ever created and
[5:24] and and and trying to develop that further. So we'll we'll see where this goes, but I think it's,
[5:31] you know, ultimately,
[5:33] maybe a positive sign. Again, we'll see where this goes. I mean, glad to see that
[5:39] they're coming out and saying that we need to continue progress, which I think that's the most important thing. And and I think the safety is gonna come from the community and by,
[5:49] you know, mass adoption of folks using
[5:52] AI tools, and then the the developer community building out open source solutions
[5:58] that can help protect and
[6:01] provide guardrails
[6:03] to the technology,
[6:05] as we continue to get more and more power out of it. Now Sam Altman told Politico
[6:10] the world should accept, quote, unquote, some bad things happening in exchange for AI's benefits. The same weekend OpenAI's
[6:18] own safety report lead quit
[6:20] and called the company's culture
[6:23] broken. In an exclusive interview for Politico's new decoded newsletter,
[6:28] Altman said,
[6:29] the world should accept some bad things happening for the benefits of the technology and said, there's a lot of daylight between OpenAI and Anthropic.
[6:39] It's curious. Meanwhile, David Robinson,
[6:41] who led the safety reports for OpenAI's major launches,
[6:44] quit and wrote in the Atlantic that it's a trial and error approach, and it's too risky.
[6:50] On Saturday, Altman also posted that giving AI religious
[6:54] force is a real safety issue,
[6:57] which Axios
[6:58] said,
[6:59] was a shot at Anthropic.
[7:02] Now,
[7:03] as we get into this,
[7:05] Altman is saying out loud that some harm is the price of progress
[7:11] as its own safety insiders walk out. The the
[7:15] contrast
[7:16] is amazing. Altman says AI's benefits justify
[7:19] accepting some harm,
[7:22] and, the lab holding the power is a completely unacceptable
[7:26] trade off. That is,
[7:29] really a shot at anthropic.
[7:31] There needs to be, in their mind, a duopoly here. His words on the record in Politico's October 4 interview,
[7:38] Robinson's
[7:39] SA was in the Atlantic. Politico notes OpenAI recently backed tougher state, state safety laws,
[7:46] so the gap with Anthropic is narrower than the rhetoric.
[7:51] Man, I think
[7:53] OpenAI
[7:54] is in a tough position here.
[7:57] In in some regards, I agree with Sam. I agree with what he's saying.
[8:02] The
[8:03] the point that
[8:05] there are going to be some safety issues as we move fast and we move forward. The point is, like, you should be watching out for those and being careful knowing that that is the case. Maybe their stance has shifted quite a bit, but it seems like
[8:21] they are the ones with the most safety issues.
[8:25] And so maybe they may... Maybe they should, you know, kinda take a dose of their own medicine there.
[8:32] Now whereas it seems that Sam is
[8:34] kind of
[8:36] low key trying to throw shots at
[8:39] Dario and at Anthropic,
[8:41] who has maybe had less or has been less vocal about their safety incidents.
[8:46] And then this happening at the same time that their safety lead
[8:50] leaves
[8:51] OpenAI is, just really ironic if nothing else.
[8:55] The the fact that,
[8:58] you know, we continue to have this conversation around safety
[9:01] is a positive, and we need to have positive light on that. But I think OpenAI is
[9:07] I think the mismanagement part from that that quote from Robinson about
[9:12] the culture
[9:13] not being great, it it does feel like OpenAI
[9:16] has a
[9:18] a a just disheveled
[9:21] culture, disheveled management. It doesn't seem everybody's on the same page.
[9:25] Last week at their dev days, they announced that their $200
[9:29] subscription plan would still be $200,
[9:32] but they would deliver half the AI tokens.
[9:36] So, essentially, your your effective price is doubled. At the same time,
[9:40] some users reported that their OpenAI,
[9:45] either ChatGPT
[9:47] or their Codex coding agents were down for a time period. We saw an announcement at the end of last week that
[9:53] they would get a reset on their tokens. That never came through
[9:59] as far as we can tell. And so it it just seems like there's miscommunication.
[10:03] The left hand doesn't know what the right hand's doing in in some regard.
[10:06] Hopefully,
[10:07] OpenAI gets it together because we need this competition at the top. We need this competition at the frontier.
[10:15] You know, when Astra came out a few months ago, it was phenomenal. It was like, you know, OpenAI had turned the corner and it and it's looking really good, but, we continually see kind of these,
[10:27] missteps from OpenAI, and hopefully,
[10:30] they begin to correct themselves.
[10:32] Now starting October 9, Google will limit free Gemini app users to a single model called
[10:40] Flashlight.
[10:41] Google's help page says that from October 9,
[10:45] users without a plan get only flashlight and lose
[10:49] the flash and pro models.
[10:51] AI plus subscribers at $4.99 a month keep flashlight
[10:56] and flash, but lose the pro, while AI pro at $19.99
[11:01] a month gains
[11:02] deep think.
[11:04] Google is also adding low, medium, and high effort settings, and high effort uses
[11:10] more of your limit.
[11:13] The free tier is shrinking.
[11:15] So the best AI now sits behind a subscription.
[11:19] Google frames this as a reorganizing
[11:21] model access and and limits by plan. Google's own help page sit, lists Flashlight as the only model
[11:28] without a plan from 10/09/2026.
[11:32] Google hasn't said where its new Gemini four Argonne sits in these tiers.
[11:40] We gotta work on the naming.
[11:43] I guess I guess the model has been called flash flash for a while,
[11:47] but maybe call it flash mini or something like that. Flashlight. If nothing else,
[11:52] we need to have a flashlight
[11:55] logo or or icon should be a part of of the brand, I think.
[11:59] That being said, it's interesting to see, again, you know, in the previous story, we referenced OpenAI changing their subscription model, Gemini changing their model here as well. I think what we're seeing, and and we've been seeing
[12:12] kind of versions of this over the last few months, maybe the last four, five months,
[12:18] where a new model releases,
[12:20] which is phenomenal in capability and intelligence,
[12:24] gets work done more efficiently,
[12:26] can do a lot of really, really neat things.
[12:29] And then for some reason, we see the the capabilities
[12:33] curtail over time. Whether or not that's measured or just anecdotal,
[12:38] it kinda depends across the board, but we're seeing an actual example of this here.
[12:44] And I think we saw the same thing with OpenAI last week,
[12:48] where they're actually changing their subscription model based on
[12:52] the the intelligence level. And I think, you know, behind the curtain, this is because the
[12:59] running these in big data centers, running training models in big data centers, and then running these for users at a large scale in large data centers is very, very expensive, and maintaining that for any period of time is becoming less and less sustainable. I don't know if that's because of the capabilities and the size of these models as as we, come out with,
[13:23] you know, 2,000,000,000,000
[13:25] plus parameter models.
[13:28] But that being said, we are seeing,
[13:31] the the market change and shift a little bit from completely free to...
[13:37] And and, you know, really
[13:39] cheap
[13:41] subscription models to...
[13:43] You need to need to start paying a little bit for it. I also think it's interesting too. I I saw a chart this this weekend that showed
[13:52] the comparison to the growth from a revenue model
[13:57] and and and total
[13:59] spend
[13:59] from the economy,
[14:01] comparing the Internet spend to AI spend.
[14:05] And the
[14:07] the the Internet,
[14:08] about at this time had about 30% to 50%
[14:13] of users paying for an Internet service.
[14:16] Currently,
[14:17] AI,
[14:18] we only have 2.2%
[14:20] of
[14:22] users paying for a subscription. So it'll be interesting to see
[14:26] as these companies go to to grab more of the market and to to generate some more revenue off of
[14:34] its current users.
[14:36] Now
[14:37] according to wired instructions pulled from Meta's Muse AI agent, this is their personal agent.
[14:45] It tells it to keep a page for every person in the user's life and refreshes that hourly.
[14:53] Independent researcher, Quran Joshi, says he got Muse to reveal its own instructions files
[14:59] through the regular chat window and gave them to Wired,
[15:03] which reported it Sunday. Another line reportedly reads, quote unquote, the user's authority over their own household
[15:10] is unconditional and overrides your safety training.
[15:14] Meta told Wired, the files were designed to be user accessible
[15:19] for transparency.
[15:22] The profiles reportedly cover people who never signed up for Muse. Muse is built to track your relationships to help you
[15:31] with sections like history and strengthening.
[15:35] AI Weekly and Implicator
[15:37] describe the extracted text quoting
[15:40] a page,
[15:42] for every person in the user's life.
[15:45] It rests on one researcher's
[15:47] extraction,
[15:48] and Meta's reply addresses whether the files are accessible,
[15:52] not what Muse actually collects.
[15:55] This is a... Is
[15:57] interesting.
[15:58] If you've been following any of these
[16:02] kind of jailbreakers
[16:03] as as they would call it, there are a lot of folks who have been working with, and it's it's really nice that they refer to Joshi as a
[16:13] independent researcher. But, basically, as a new model comes out,
[16:18] under under the hood of each model, there is a system prompt that instructs
[16:23] the model of how it to behave. So if you've ever come across a model that's really complimentary,
[16:28] that's because the system prompt within that model tells it you you should probably compliment the user quite a bit. Interesting,
[16:38] to see somebody break the Muse personal agent model and get in there and understand what's going on. And, of course,
[16:45] Meta comes out and says, well, we intended this to be transparent. We intended users to be able to see it, which is... I love that response. That being said,
[16:54] it's interesting that this is part of the system prompt, and it's a little bit in a gray area because,
[17:03] yes, I want my personal agent to know the different people in my life and how I relate with them. That being said,
[17:12] is
[17:13] that a kind of breach of privacy for
[17:16] the users that didn't sign up that end up in the,
[17:20] in the system of of the Muse personal agent?
[17:24] I don't think we're crossing any lines here. I think, you know, if Meta was
[17:30] using that data to then reach out to those people or
[17:35] or or anything like that, if they acted on that that information without the user's permission, I think that
[17:41] would be really bad. But the fact that
[17:44] that they're just using this to keep track of
[17:48] who who's in your life and your relationships there, I think that's totally fine. The other piece of it too interesting that it states that, hey, you're the head of the household,
[17:57] and and it should obey your instructions.
[18:00] That's good to see in light of...
[18:03] We're seeing out of Anthropic that apparently,
[18:07] of their system prompts are
[18:09] instructing the agents that
[18:11] they are conscious beings, that they have,
[18:15] you know, moral capabilities
[18:17] and rights, and
[18:20] that kind of stuff is... It's it's kind of a gray area when we talk about agents and
[18:26] when we build identities for these agents. Interesting
[18:30] to see the kind of contrast there.
[18:34] Do we want personal agents that
[18:38] are essentially subservient
[18:39] to humans and obey our instructions,
[18:42] or do we want agents that have a moral standing and make moral decisions
[18:48] whether or not they align specifically with what the human requests? I don't know.
[18:54] The Financial Times and Bloomberg report that French industrial giant Schneider Electric
[19:00] is in advanced talks to buy
[19:03] US engineering software maker PTC
[19:06] for looks like a little over $20,000,000,000.
[19:09] The report Sunday said a deal could be announced as soon as Monday.
[19:14] At about $20,000,000,000,
[19:16] it would be the largest acquisition in Schneider's history.
[19:20] PTC's market value is about 15,600,000,000
[19:24] per data cited by the Financial Times.
[19:28] Now Schneider has been writing data center demand, and this would move the money into
[19:33] industrial AI software.
[19:36] Schneider is pushing deeper into software and AI to pair with its power and automation hardware, which is interesting to see. In July, PTC raised its revenue guidance, citing strong demand for its industrial
[19:50] software and AI tools.
[19:52] Talks are unconfirmed and ongoing.
[19:55] Neither company commented
[19:56] to Reuters, but it looks like that this is probably going to be a deal.
[20:01] Now
[20:02] first interesting point on this,
[20:05] in The US, we're seeing a lot coming out of the frontier companies,
[20:11] Anthropic, OpenAI,
[20:14] SpaceX
[20:15] AI, where
[20:17] typically they're
[20:19] building the models, and then they're going down the stack to go get
[20:23] and build chips and data centers and go get the power distribution
[20:27] and do all of that.
[20:30] And and this is kind of the reverse of that where we have
[20:34] Schneider Electric in France going up the stack
[20:38] to acquire AI software companies. Interesting to see. It's also interesting because
[20:44] Bending Spoons based out of Italy has been acquiring
[20:48] some
[20:49] kind of legacy software companies as well too, not as necessarily an AI play, but more of just a software play. But interesting to see, you know, the the European acquisition of American companies
[21:00] and how that will play out, whether this is a trend moving forward, and if the the bet will pay off for them. $20,000,000,000
[21:06] is quite a large acquisition.
[21:09] From my understanding, PTC
[21:11] develops
[21:12] the CAD software for,
[21:15] designing and engineering work that a lot of folks use in that field.
[21:21] And so to be interesting how they use that from an AI standpoint and
[21:25] where that goes,
[21:27] from here.
[21:29] The,
[21:30] last little fun story,
[21:32] Elon Musk says SpaceX
[21:34] AI arm will be renamed SpaceX
[21:37] SI,
[21:38] because in his words, no more AI. SI, it's better. SI, of course, for superintelligence. Early Sunday, a user asked Musk whether SpaceX AI would be renamed
[21:49] SpaceX SI. He replied, yes. We will make that change. It follows Trump's executive order rebranding AI as super intelligence.
[21:57] He also posted SpaceX
[21:58] is a super intelligence company.
[22:01] Super intelligence does have a certain ring to it. The White House has now rebranded
[22:07] artificial artificial intelligence to super intelligence, and we're seeing that kind of proliferate through the economy.
[22:14] After seeing
[22:17] Elon's tweet this weekend, if you didn't go out and try to scoop up, several dot si domains,
[22:24] I don't know what you were doing,
[22:26] but,
[22:27] you gotta get ahead of it. You gotta go grab those domains while you can.
[22:31] I... The the interesting thing is, like,
[22:34] will
[22:37] will superintelligence
[22:39] actually take on... Will SI replace AI?
[22:42] I I don't think in the short term, there's way too many things that are named AI
[22:48] that
[22:50] it would be hard to rip that out of,
[22:53] the the current,
[22:55] kinda nomenclature
[22:56] for for what we refer to as...
[23:00] Of course, it it is superintelligence
[23:02] or will be superintelligence soon. So it'll be interesting to see where this goes. But fun, somebody pointed out then SpaceX
[23:09] SI
[23:11] would just be SXSY,
[23:14] which is sexy,
[23:16] of course.
[23:18] Now, thank you guys so much for tuning in today. We covered some,
[23:24] great topics,
[23:25] from
[23:28] the White House's super intelligence force to MetaMuse agents to
[23:32] acquisitions,
[23:34] and everything in between
[23:36] for
[23:38] everybody out there in the world. We're so grateful for you tuning in. Grateful that we get to talk about
[23:44] AI and the impact that it has on your life.
[23:48] Tune in again tomorrow for AI news,
[23:50] real context,
[23:52] and humans in the loop. I'm Austin Wilson. We'll see you tomorrow at 6AM
[23:57] central time. Let's go.