2026-06-01

Opus 4.8 Drops, Demis Hassabis Predicts AGI, and the $220B Foundation | EP #260

The panel covers Anthropic's Opus 4.8 reclaiming the coding-benchmark lead from GPT-5.5 and debates whether frontier-model benchmarks are saturating. They dig into Demis Hassabis tightening his AGI timeline to 2029 and reignite the panel's long-running argument over what AGI even means. Other stories include Amazon turning its Alexa shopping assistant into an AWS-style platform for retailers, the eye-popping size and mandate of the new OpenAI Foundation (26% of OpenAI, $130-260B), IBM and the US government's $2B Anderon quantum chip foundry, wind and solar surpassing natural gas globally for the first time, a $5 handheld cancer-detecting blood sensor out of China, and a sober look at anti-tech extremism being classified as a domestic security threat. The episode closes with a rapid-fire audience Q&A on UBI/UBS models, privacy-as-agency, token cost collapse, and Alex's speculative case for computing with gravity.

▶ Watch on YouTube

Topics

Opus 4.8 reclaims coding crown AI ▶ 3:39

Anthropic released Opus 4.8 six weeks after 4.7, topping the Artificial Analysis Intelligence Index (61.4) and SWE-Bench Pro (69.2% vs GPT-5.5's 58.6%), and is 4x less likely to overlook its own bugs.

Benchmark saturation and the need for new evals AI ▶ 4:34

Alex argues frontier labs are converging because compute levels are within 2-3x of each other and existing benchmarks (SWE-Bench Pro, HLE, GDPval) are saturating; he calls for new benchmarks built around unsolved scientific/engineering problems.

Claude Code dynamic workflows and self-forking agents AI ▶ 6:16

Anthropic shipped a feature letting Claude Code spin up hundreds of parallel sub-agents; Dave describes running ~100 agents on EC2 and a new 'self-fork' capability that clones an agent's full context instead of starting children from scratch.

Demis Hassabis tightens AGI timeline to 2029 AI ▶ 12:23

Hassabis aligned with Ray Kurzweil's 2029 AGI timeline, called today's agents 'a practice run,' and proposed an 'Einstein test' (can a model trained only through 1901 derive special relativity). Alex calls the timeline conveniently conservative given Gemini isn't currently leading.

What counts as AGI (panel disagreement) AI ▶ 13:28

Alex reiterates his view that some form of AGI has existed since GPT-2/2020, contrasting with Hassabis's stricter Einstein-test bar; Salim argues the AGI debate itself is a distraction ('noise') from solving concrete problems.

Amazon vs Google's agentic commerce strategies Economy ▶ 23:03

Amazon is licensing its Alexa AI shopping assistant (3.5x conversion vs keyword search) to all retailers, an AWS-style vertical play, while Google pushes an open horizontal protocol layer (Universal Cart, Universal Commerce Protocol, Agent Payment Protocol) for agentic shopping.

OpenAI Foundation becomes the world's largest philanthropy Economy ▶ 30:25

After OpenAI's PBC restructuring, the OpenAI Foundation owns 26% of OpenAI (valued $130-260B), surpassing Novo Nordisk, Tata Trusts, and the Gates Foundation. It just issued a $250M grant on economic futures (public wealth funds, worker ownership, AI dividends) and controls OpenAI's board.

UBI/UBS/UBC debate and 'privatized socialism' Economy ▶ 37:39

The panel debates whether frontier-lab foundations will eventually be pressured to fund UBI/UBS/UBC (universal basic compute/capability), with disagreement over whether this constitutes libertarianism, technological socialism, Fordism, or 'privatized socialism.'

IBM/DOE Anderon quantum chip foundry Compute ▶ 47:07

IBM and the Department of Commerce announced Anderon, a $2B purpose-built quantum chip foundry in Albany, NY, aiming to make IBM a 'TSMC of quantum' for Google, IonQ, Rigetti, and D-Wave; framed as a hedge for quantum-accelerated AI advances.

Wind and solar surpass natural gas globally Energy ▶ 51:37

In April 2026, wind and solar hit 22% of global electricity generation, surpassing natural gas's 20%, driven by exponential solar cost/capacity curves; the panel contrasts this with repeatedly wrong IEA forecasts.

Anti-tech extremism declared a domestic security threat Geopolitics ▶ 60:00

Federal agencies created a new 'anti-tech extremism' threat category after attacks on Sam Altman, monitoring threats against data centers and executives; Dave cites evidence of Chinese-funded involvement echoing Cold War-era protest tactics.

California's AI workforce-impact executive order Economy ▶ 67:16

Governor Newsom signed an order to build a public dashboard tracking AI-related job losses and vulnerable industries in real time; Dave notes only ~300,000 jobs lost to AI so far, calling it a hiring freeze rather than mass layoffs.

$5 handheld blood test detects lung cancer Biotech ▶ 72:51

West Lake University (China) built a metamaterial-based optical sensor detecting early-stage lung cancer from a single blood drop at ~95% accuracy, 10,000x more sensitive than standard labs, published in Nature Photonics, costing about $5.

US robotics wake-up call vs China's humanoid buildout Robotics ▶ 79:14

Andreessen Horowitz warned the US needs a robotics wake-up call as China executes its 'AI Plus' plan with 150+ humanoid robotics companies; Alex (who co-founded Prorata/Pro RL for robotics) and Dave describe pivoting venture capital from pure software AI toward hardware/robotics.

Predictions made

open Alex Wissner-Gross: The public will keep moving the AGI goalpost until finally declaring 'AGI sentience' as the next milestone once current AGI thresholds are met.

EP #? · · due: unspecified · ▶ watch

“We're going to keep moving the goalpost on AGI AGI AGI and then we're going to go oh AGI sentience. That's what's going to happen.”

Your call:

open Alex Wissner-Gross: By the time the Anderon quantum foundry is built out and at scale, the US will have achieved quantum-accelerated AI training or inference advances, justifying having that infrastructure onshore rather than overseas.

EP #? · · due: late 2020s · ▶ watch

“I'm pretty optimistic that sometime in the next few years, probably by the time this foundry is ready and at scale, we will have quantum accelerated AI advances.”

Your call:

open Alex Wissner-Gross: SpaceX (or a merged SpaceX/Tesla entity) will end up forced to build sun-synchronous orbital or lunar solar-panel production at scale, importing cheap Chinese panels in the near term while relying on Tesla's upstate New York facility, rather than Tesla independently solving this.

EP #? · · due: unspecified (next several years) · ▶ watch

“I expect Elon's SpaceX or SpaceX plus entity to end up being forced to do that and not Tesla, which was never a perfect fit.”

Your call:

open Dave Blundin: The window for purely software-based AI startups to generate outsized venture returns will close in roughly two years as the AI self-improvement loop takes over; robotics will be a good investment theme for about a decade.

EP #? · · due: ~2028 · ▶ watch

“The window of opportunity for pure software AI is probably another couple years and then the self-improvement loop is going to take over... I think robotics is a good 10-year theme.”

Your call:

open Alex Wissner-Gross: Non-invasive wearable optical cancer detection (extending beyond single-drop-of-blood metamaterial sensors) will become achievable.

EP #? · · due: ~5 years (roughly 2031) · ▶ watch

“I'll make a forecast... maybe say next 5 years or so. I think we will get to non-invasive wearable optical cancer detection.”

Your call:

open Alex Wissner-Gross: AI-driven advances in physics will let computing transcend semiconductors entirely within about a decade, ultimately including highly speculative substrates like gravity-based computing within roughly 10-20 years.

EP #? · · due: ~2036 (semiconductor transcendence); 10-20 years for gravity computing · ▶ watch

“I do think it is very likely that we will ultimately transcend semiconductors and CMOS, and by ultimately I mean on the time scale of a decade, not like centuries.”

Your call:

open Alex Wissner-Gross: AI agent/token costs will keep collapsing so that a $20/month AI agent service today will cost about $2/month by 2028 and $0.20/month by 2030, with free tiers also emerging from hyperscalers.

EP #? · · due: 2030 · ▶ watch

“A useful AI agent today you can buy it at 20 bucks a month will end up costing you two bucks a month in 2028 and 20 cents in 2030.”

Your call:

open Alex Wissner-Gross: Someone will publicly ask why frontier-lab nonprofit foundations (owning ~20-25% of labs asymptotically converging on a large share of world GDP) shouldn't themselves fund UBI/UBS, and this pressure will become effectively irresistible.

EP #? · · due: unspecified · ▶ watch

“Someone somewhere I'll make a prediction is going to be asking the question... why isn't the foundation itself supporting UBI or UBS basically.”

Your call:

Numbers that matter

Opus 4.8 scores 61.4 on the Artificial Analysis Intelligence Index, 1.2 points ahead of GPT-5.5 Anthropic's new flagship model reclaiming the coding/intelligence benchmark lead
Opus 4.8 scores 69.2 on SWE-Bench Pro vs GPT-5.5's 58.6 Hard coding benchmark comparison
Opus 4.8 scores 57.9% on Humanity's Last Exam with tools Panel previously set 50% on HLE as their informal internal AGI-proxy threshold
GDPval score of 1890 for current frontier models Alex's cited economic-task benchmark
Amazon's Alexa AI shopping assistant converts shoppers at 3.5x the rate of traditional keyword search Basis for Amazon licensing it to all retailers
Amazon once took 60% of product search away from Google Historical context for Alexa/product-search competitive dynamics
OpenAI Foundation owns 26% of OpenAI's PBC, valued at $130-260 billion Makes it the largest nonprofit foundation in the world
Comparable major foundations: Novo Nordisk Foundation $150B, Tata Trusts $100B, Gates Foundation $75B Comparison set showing OpenAI Foundation now exceeds all of them
OpenAI's People First AI Fund distributed $40 million to ~28 nonprofits (launched 2025) First of OpenAI Foundation's grant rounds
OpenAI gave a large grant (stated in transcript as '$25 billion,' likely a caption error for $25 million) across health breakthroughs and AI resilience in October Second major OpenAI Foundation grant round; figure may be a transcription/caption error
New $250 million OpenAI Foundation grant on economic futures research Funds research on public wealth funds, worker ownership models, and AI dividends
OpenAI's original Superalignment plan allocated 20% of OpenAI's compute to safety Alex's numerology comparison to the 26% foundation stake
US Social Security is approximately 22% of the federal budget Comparative scale reference for foundation-funded UBI discussion
Anderon quantum chip foundry is a $2 billion project ($1B CHIPS Act funds, $1B from IBM) Located in Albany, NY, using a 300mm manufacturing process
Anderon can produce quantum chip devices 30x faster than current methods IBM positioning itself as the 'TSMC of quantum'

Worth digging into

🕳️ OpenAI Foundation's governance and $130-260B valuation

It's now the largest philanthropic foundation in the world, controls the OpenAI PBC board with only 26% ownership, and the panel speculates it may eventually be pressured into funding UBI/UBS directly.

🕳️ IBM/DOE Anderon quantum chip foundry

A $2B bet framed as making IBM the 'TSMC of quantum,' timed to be ready roughly when quantum-accelerated AI advances are expected.

🕳️ China's 150+ humanoid robotics companies under the 'AI Plus' plan

Andreessen Horowitz is warning this repeats the solar/5G pattern where China achieved industrial dominance while the US fell behind; the panel treats it as an active VC thesis shift.

🕳️ West Lake University's $5 metamaterial blood-cancer sensor

A striking demonetization claim (95% accuracy, 10,000x lab sensitivity, $5 cost) explicitly positioned as succeeding where Theranos failed, with echoes of Verily's abandoned smartwatch cancer-detection project.

🕳️ Alex Wissner-Gross's 'computing with gravity' claim

A highly speculative but specific physics claim (turbulent strong-gravity regimes may be Turing-complete) used to argue computing could eventually transcend semiconductors entirely.

🕳️ California's AI workforce-impact dashboard (Newsom executive order)

Framed as the most concrete government action yet on AI job displacement, and a test case for real-time labor-market sensing versus the Dallas Fed's 'hiring freeze, not mass layoffs' read.