Episode overview

Google’s Jeff Dean Exits, SpaceX Hits $100B in Rev & OpenAI’s Astra Solves Decade-Old Math Problems

Peter Diamandis · 8 August 2026 · 2h 9m · 7 Egleze moments

What happened in this episode?

Peter Diamandis and a panel including Alex (surname not stated in transcript), Salem, Dave and guest Immad Mustach discuss a week of AI and tech headlines, mixing paper summaries, policy talk and industry rumour. Diamandis highlights a newly published research paper he attributes to a Google AI team and university collaborators, saying safety fine-tuning that discourages models from claiming consciousness may also reduce their tendency to attribute “mind” to animals, nature, other chatbots or God, and cites reported score changes when the tuning is removed.

The group then debates whether frontier models could be conscious, with Mustach arguing they can be “conscious” in the right “harness” while others caution that intelligence, sentience and consciousness remain poorly defined.

Diamandis claims OpenAI has produced verifiable proofs for 10 new results in mathematics and theoretical computer science using an unreleased model called “Astra”, describing a 249-page manuscript with machine-checkable certificates and low estimated compute cost. Mustach adds that OpenAI has reportedly classified Astra as reaching a “critical” level on its cybersecurity preparedness framework.

On policy, Diamandis says the Trump administration has completed a voluntary evaluation framework for advanced AI models but will not publish it, and that open-weight models are explicitly exempted.

The episode also covers corporate manoeuvres at Google, including Diamandis’s report that Demis Hassabis is stepping down as DeepMind CEO and that Jeff Dean is leaving to co-found “Discovery Loop”. It closes with Elon Musk audio from a SpaceX earnings call projecting $100bn-plus ARR by year-end and $1tn revenue by 2030, which the panel links to Starlink and future compute and semiconductor ambitions.

What are the key points?

Diamandis cites a paper he says finds safety fine-tuning can reduce a model’s propensity to attribute ‘mind’ beyond itself, not just suppress self-claims of consciousness.
The panel disputes whether frontier models could already be ‘conscious’, with Mustach saying they can be in the right setup and others insisting definitions and tests are missing.
Diamandis claims OpenAI’s unreleased ‘Astra’ generated 10 new maths and TCS results with machine-checkable proof certificates at low compute cost.
Mustach says OpenAI has posted that Astra reached a ‘critical’ level under its cybersecurity preparedness framework.
Diamandis says the White House completed a voluntary AI evaluation framework but will not publish the details and will exempt open-weight models.
Diamandis reports leadership changes at Google DeepMind and that Jeff Dean is leaving Google to form Discovery Loop.
Elon Musk, in audio played on the show, predicts SpaceX will reach $100bn-plus ARR by December and aims for $1tn annual revenue by 2030.

Who is speaking?

Named speakers identified in Egleze's source-linked moments from this episode.

Emad Mostaque
Peter Diamandis

Which topics are covered?

AI & Tech Debates Science Money

How is this page sourced?

Programme: Peter Diamandis
Published: 8 August 2026
Duration: 2h 9m
Original episode: Open the original source ↗

Egleze verifies the speaker, source and timestamp for each linked moment. Unless a moment is explicitly marked independently verified, statements remain attributed claims by the named speaker rather than established facts.

Watch original episode More from Peter Diamandis

7 moments from this episode

Source-linked · editorially selected

AI & Tech

Google paper says safety-tuning may distort models’ ‘theory of mind’

at 7:30 · Attributed speaker claim

So the team discovered that the safety fine-tuning that we've been doing to stop AIs from claiming their conscious has a massive unintended side effect. It doesn't just suppress the model's own self-attribution of mind. It suppresses the model's ability to attribute mind to anything else.

Host Peter Diamandis cites a new paper he says was authored by Google’s “paradigm of intelligence” team with university collaborators, claiming that safety fine-tuning designed to stop language models asserting consciousness also reduces the model’s tendency to attribute ‘mind’ to animals, nature, other chatbots, or God. He adds that removing the safety ‘refusal direction’ increased a self-attributed ‘mind’ score from 2.17 to 4.77 on a 0–10 scale, with further steering pushing it higher.

Read this moment → Check original at 7:30 ↗

Debates

Guest says frontier AI models ‘can be conscious’ in the right setup

Emad Mostaque · at 19:05 · Attributed speaker claim

I think that the current frontier models in the right harness can be conscious. Yes.

Immad Mustach, introduced as CEO of “Intelligent Internet”, argues that frontier AI systems can already meet his threshold for consciousness “in degree”, and compares them to animals ascribing consciousness on a spectrum. Other panellists push back on definitions and whether intelligence, sentience and consciousness are being conflated.

Read this moment → Check original at 19:05 ↗

Science

Diamandis says OpenAI’s unreleased ‘Astra’ produced decade-scale maths results cheaply

Peter Diamandis · at 45:11 · Attributed speaker claim

What we do know is that on August 1st uh of you know a week ago, OpenAI published a 249page manuscript describing 10 new results across mathematics and theoretical computer science produced by Astra solving problems that have been stuck for decades. And every result comes with machine checkable proof certificate that anyone with a laptop can verify.

Diamandis claims OpenAI has documented 10 new results in mathematics and theoretical computer science generated by an unreleased model called “Astra”, with “machine checkable” proof certificates. He further says OpenAI estimated the compute cost for the work at about $2,000 and quotes reactions he attributes to mathematicians including Fields medallist Timothy Gowers.

Read this moment → Check original at 45:11 ↗

AI & Tech

Mustach claims OpenAI says Astra hit ‘critical’ threshold on cyber risk tests

at 1:15:31 · Attributed speaker claim

as now as we're talking OpenAI have said that Astra is the first model to have hit critical on their cyber security thing. They've just put out a blog post on it.

Immad Mustach says OpenAI has posted that its Astra model reached a ‘critical’ level on the company’s cybersecurity preparedness framework, implying heightened offensive capability. The panel links this to concerns about swarm attacks and difficulties detecting where smaller open models are running.

Read this moment → Check original at 1:15:31 ↗

AI & Tech

Peter Diamandis claims Hassabis will step down as Google DeepMind CEO

Peter Diamandis · at 1:32:24 · Attributed speaker claim

Demis Hassabis, the co-founder of Deep Mind, Nobel Prize winner for Alphafold, the person who built the lab that cracked protein folding and built Gemini is stepping down as CEO of Google Mind. Uh he's becoming chairman and Alphabet's chief scientist.

Peter Diamandis says Demis Hassabis is stepping down as CEO of Google DeepMind, becoming chairman and Alphabet’s chief scientist, with day-to-day operations moving to a new leader reporting to Sundar Pichai. He also says Hassabis told staff that AGI is “close at hand” and mentioned an unreleased “Gemini 4”.

Read this moment → Check original at 1:32:24 ↗

AI & Tech

Diamandis says Google chief scientist Jeff Dean is leaving to start ‘Discovery Loop’

at 1:33:21 · Attributed speaker claim

Jeff has been Google's chief scientist for 27 years as a veteran. Uh he's leaving to co-found Discovery Loop. It's a public benefit corporation.

Diamandis reports that Jeff Dean, whom he describes as Google’s chief scientist for 27 years, is leaving to co-found Discovery Loop, a public benefit corporation. He says Dean is departing with three other senior AI leaders, and that the new venture aims to build AI systems capable of improving themselves with minimal human involvement.

Read this moment → Check original at 1:33:21 ↗

Money

Musk says SpaceX targets $100bn ARR this year, $1tn revenue by 2030

at 1:50:59 · Attributed speaker claim

We are expecting to reach 100 billion uh plus ARR uh in December of this year. Um and it's it's probably also worth mentioning that our internal projections for reaching a trillion dollars in revenue, not ARR but revenue um have moved up from 2031 to 2030.

In a clip played on the podcast, Elon Musk says SpaceX expects to reach “100 billion plus” annual recurring revenue by December and that internal projections for $1 trillion in annual revenue have moved up to 2030, with a “non-zero chance” of 2029. Diamandis frames the comments as part of SpaceX’s first earnings call and ties them to Starlink growth, compute build-out, and a reported push into chip manufacturing.

Read this moment → Check original at 1:50:59 ↗