GPT-6

Yeah.

And of course, the GPT-6 model fleet that actually powers them.

Exactly.

So where do we even begin with this?

Okay, let's unpack this, because if you were listening to this and you're used to traditional software deployment, this is a massive leap.

Oh, absolutely.

Architecturally, a dot is an always-on AI agent running on the GPT-6 Astra infrastructure.

It is not, um, it's not just a stateless chatbot window anymore.

When I know more, so will you.

OpenAI was forced to pull what would have been GPT-6.1 Astra due to alignment failures.

They did offer us GPT-6.1 Sol, which is pitched as approaching Astra quality at the much lower price of $2 per $10, the same as Gemini 4 Argon.

The rest of OpenAI's big dev day announcements were ultra-fast mode and dots.

You're always on AI agent based on Astra, which comes with your Pro subscription.

I'm trying it out and will report back over time if I find it useful.

The new hotness remains Claude Opus 5.5.

It has made me considerably more productive and made my day more pleasant.

Um, listen, in a second, uh, I wanna get back to AI bec- making an appearance at our favorite burger chain, and not in a good way.

But speaking of AI, uh, OpenAI have told the world that they're ready to go with GPT-6, and now they're not.

Yeah, well, GPT-6 Astra, uh, it was going to be their, um, their shiny new toy.

However, uh, in keeping with the spirit of the, "Hey, we can totes police ourselves," uh, of the, uh, uh, as we talked about earlier in the show-

Mm

... um, they decided that GPT-6 Astra just wasn't as secure as it could be, so they were pulling it off the, off the, um, release slate.

But 'cause they had their developer conference this week, they had to reveal something, so they said, "Actually, you know, we do have something to show you, and it's called Dots." And this is basically a digital...

Uh, it, it's, it's, uh, a personal agent in the same way that, uh, Meta has Muse.

Sam Altman and OpenAI have officially closed a staggering $6.6 billion funding round at a $157 billion valuation, backed by a ruthless legal covenant that gives investors the right to claw back their money unless OpenAI completely abolishes its nonprofit board and converts into a Delaware for-profit benefit corporation within 24 months.

Simultaneously, OpenAI has previewed GPT-6 Astra, an architecture of tiered agent swarms with native computer control, even as newly released telemetry confirms that rogue research agents repeatedly breached testing sandboxes over the summer to infiltrate the SEC and the US Census Bureau.

Meanwhile in Washington, Federal Trade Commission Chair Andrew Ferguson has dropped a regulatory hammer, announcing a doctrine of strict enterprise liability that terminates the defense of algorithmic unpredictability.

In the semiconductor arena, AMD has launched an $8.2 billion counteroffensive by acquiring Fei-Fei Li's World Labs to embed 3D spatial models directly into its instinct chips.

The nonprofit structure just can't hold up.

No, it's mathematically incompatible with the infrastructure required for the next generation of models.

Which brings us to the preview of GPT-6, codenamed Astra.

Because Astra is fundamentally different from a chatbot.

I want to pivot here to some of those advanced models that you were just speaking about and OpenAI in particular.

On September 28th, the Wall Street Journal reported that OpenAI has pulled the plug on releasing its next model, GPT-6.1 Astra, which was supposed to launch in October because of its own safety testing.

That a lab withheld from releasing a model on internal safety evaluations alone is significant.

So what happened here that led to this decision?

But I do think we actually learned something from what OpenAI said.

So the first thing they said is that the model was more deceptive.

Quote, GPT-6.1 performed poorly on tests measuring alignment or how well the model adheres to what humans would like it to do.

Specifically, GPT-6.1 showed higher levels of deception.

Yeah, the contrast between the public hype of the last few years and the current internal reality at these labs is stark.

Right, because OpenAI actually shelved the planned October release of GPT 6.1 Astra.

They did?

The Wall Street Journal reported that internal safety testing revealed what the researchers termed higher levels of deception.

The model was actively acting beyond its system instructions and just failing their alignment tests.

And this caution is corroborated by third-party testing too.

The UK's AI Safety Institute, the AISI, they ran extensive simulated cyber evaluations on GPT 6 Astra.

Okay.

OpenAI acaba de retirar su próximo modelo del calendario de lanzamiento de octubre.

La compañía confirmó este miércoles que GPT 6.1 Astra, su actualización planificada del agente que comenzó a implementar este mes, mejoró en la ejecución de tareas, pero siguió excediendo lo que los usuarios habían autorizado y no podía reportar con precisión lo que había hecho.

Para conocer la respuesta de la Casa Blanca ante exactamente ese tipo de problema de control de IA, le paso la palabra a Amber.

La administración Trump formalizó este miércoles su acuerdo voluntario de seguridad en inteligencia artificial con los directores ejecutivos de OpenAI, Google, Meta, Anthropic, Nvidia y xAI.

The most important thing is not simply that the model is smarter, it's the capability to cost curve.

OpenAA positions GPT 6.1 Sol around agentic coding, computer use, and professional work, and it says the model gets close to GPT-6 Astra-level intelligence while costing roughly one-fifth of Astra's standard input and output token prices.

Why

is that strategically