Explore

May 2026

Demis Hassabis on his rush to 'solve all disease' and Isomorphic's new $2.1 billion (Fortune)

Hassabis discusses Isomorphic Labs' fresh $2.1B Series B (led by Thrive Capital), arguing its near-term bottleneck is compute, not the AGI-scale compute of DeepMind, so more compute means more...

Import AI 458: Reckoning with the future; and a singularity story

Clark argues AI progress will continue predictably with compute/data scaling and that recursive self-improvement (an AI building its own successor) could plausibly arrive within two years, framing...

Sam Altman: 'We try to think out loud' — Accelerate AI conversation with CBA CEO Matt Comyn

Altman argues AI capability has reached "a notable place" but economic and enterprise adoption is "still very early," creating a leadership gap that demands transparency about uncertainty rather than...

Google DeepMind CEO Demis Hassabis says we're close to AGI (Axios interview after Google I/O)

Hassabis says we're in the "foothills of the singularity" and expects AGI around 2030, now seeing 2029 as possible, with only a few years to prepare. He frames the coming "agentic era" as a societal...

Demis Hassabis isn't shying away from AI's biggest questions (Fast Company, Google I/O interview)

Hassabis predicts AGI will arrive around 2030 (plus or minus a year) and argues Google balances safety against the most voracious competition in tech history better than anyone. He champions secure,...

DeepMind founder Demis Hassabis on what Google AI products say about 'singularity' (Semafor)

Hassabis argues we are "at the foothills of the singularity," meaning the current AGI era, citing agents starting to work, coding maturing, and science/math being accelerated as signs of the gradual...

Import AI 457: AI stuxnet; cursed Muon optimizer; and positive alignment

Clark covers four items with his own analysis: a ~20-year-old virus (fast16.sys) that subtly sabotaged high-precision engineering/physics simulation tools, which he frames as a preview of how a...

Sam Altman trial testimony in Musk v. Altman (says OpenAI was 'left for dead')

Testifying in the Musk v. Altman trial, Altman argues he did not "steal a charity" but rather that Musk abandoned OpenAI ("We were kind of left for dead") after failing to keep funding promises and...

Anthropic's Daniela Amodei says entrepreneurs should go on vacation to road test potential cofounders (Fortune)

Daniela Amodei argues the best way to vet a potential cofounder is to vacation together first ("share a room") as a low-stakes road test: if you want more time together it's a good fit, if you need...

Import AI 456: RSI and economic growth; radical optionality for AI regulation; and a neural computer

Clark reviews several AI research developments and weighs in with his own views. He endorses the Institute for Law & AI's "radical optionality" approach to AI regulation (build government...

Ex-OpenAI exec Sutskever says he spent a year gathering proof of alleged Altman dishonesty (Reuters)

Testifying in the OpenAI vs. Musk trial, Sutskever said he spent roughly a year gathering evidence at the board's request to document a "consistent pattern of lying" by Sam Altman, compiling a...

Behind the Curtain: Intelligence explosion (Axios video interview with Jack Clark)

Clark predicts a 60%+ chance that by end of 2028 an AI system could be told to "make a better version of yourself" and do so fully autonomously, citing early signs of recursive self-improvement (AI...

A conversation with Dario Amodei & Daniela Amodei (Code with Claude SF 2026)

Dario Amodei argues the AI exponential is unfolding as the scaling laws predicted over a decade ago, yet feels viscerally shocking anyway; Anthropic saw roughly 80x annualized revenue/usage growth in...

Dario Amodei spent last year warning of an AI white-collar bloodbath. Now he's changing the narrative

At Anthropic's financial-services press briefing alongside JPMorgan's Jamie Dimon, Amodei pivots from his earlier warning that AI could eliminate half of entry-level white-collar jobs toward a more...

Anthropic CEO Dario Amodei warns some software companies will 'completely go bust'

Dario Amodei argues that SaaS companies whose competitive moat rests on the complexity of their software are losing that moat as AI makes such software easy to replicate. He says individual SaaS...

OpenAI trial: Brockman rebuts Musk's take on startup's history, recounts secret work for Tesla

Concluding his testimony in the Musk v. Altman trial, Brockman rebutted Musk's account of OpenAI's founding: he said he never promised Musk anything about the corporate structure, that OpenAI...

Import AI 455: Automating AI Research

Clark argues there is a 60%+ chance that fully automated, no-human-involved AI R&D (a model capable of autonomously training its own successor) arrives by the end of 2028, calling it a Rubicon that...

OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck (Training Data podcast)

Brockman frames OpenAI's business simply as buying/renting/building compute and reselling it at a margin, says demand for intelligence is unlimited and they can never buy enough compute. He argues...

April 2026

Demis Hassabis on Building DeepMind, AlphaFold, and the Final Stretch to AGI (Training Data podcast, AI Ascent 2026)

Hassabis traces the common thread from chess and games (Theme Park, Elixir Studios' over-ambitious Republic) to founding DeepMind in 2009, arguing the key insight was combining deep learning with...

How to Build the Future: Demis Hassabis (YC podcast episode)

Hassabis argues the current paradigm (large-scale pre-training, RLHF, chain-of-thought) will be part of the final AGI architecture but is missing one or two pieces, naming continual learning,...

An Interview with OpenAI CEO Sam Altman and AWS CEO Matt Garman About Bedrock Managed Agents

Altman argues AI is the fourth great platform-enablement moment for startups (after internet, cloud, mobile), and that revenue is now scaling for young companies at an unprecedented rate. He frames...

Our Principles (OpenAI)

Altman lays out five principles guiding OpenAI's work toward AGI: democratization (resisting concentration of power, putting general AI in as many hands as possible via democratic/egalitarian...

An Interview with OpenAI President Greg Brockman: 'There's not going to be enough compute'

Brockman argues the world is heading into a "compute-powered economy" where there will never be enough compute to meet demand, citing how agentic coding exploded after December and how OpenAI has...

Live: Anthropic co-founder Jack Clark on AI and jobs (Planet Money Live)

Clark argues that if AI advances as expected, capitalism must be reconceptualized: robots and AI companies should be taxed heavily to reallocate money from the "machine economy" to the human economy...

Greg Brockman: Inside the 72 Hours That Almost Killed OpenAI (The Knowledge Project)

Brockman recounts founding OpenAI: leaving Stripe for a mission he'd dedicate his life to, the 2015 dinner with Sam Altman, and the Napa offsite where he made t-shirts and recruited researchers...

The Great Reset At OpenAI — EP 67 Sam Altman and Greg Brockman (Core Memory)

Altman reflects on OpenAI's 10-year arc and his working relationship with Brockman, and argues the field (and OpenAI itself) fell into a "tech nerd" frame of promising superintelligence without...

Import AI 454: Automating alignment research; safety study of a Chinese model; HiFloat4

Clark analyzes several AI developments with his own framing: Huawei's HiFloat4 4-bit training format beating MXFP4 on Ascend NPUs, which he reads as a symptom of export controls pushing China toward...

Sam Altman: AI in Hollywood Will Get People to Care More About Human Creators, Not Less (Breakthrough Prize)

Asked about Hollywood fears that AI will cut jobs and creativity, Altman argues the opposite will happen: because people fundamentally care about the humans behind stories and art, AI will make...

The billionaire Anthropic cofounder who majored in literature says knowing how to ask the right questions beats knowing how to code (Fortune)

Clark argues that as AI handles rote programming, the most valuable skills are knowing the right questions to ask and synthesizing insights across disciplines, crediting his English literature...

Import AI 453: Breaking AI agents; MirrorCode; and ten views on gradual disempowerment

Clark argues AI is more capable at long-horizon coding than most realize, citing METR/Epoch's MirrorCode benchmark where Claude Opus 4.6 reimplemented a 16,000-line Go toolkit a human would need...

Howard University Fireside Chat on AI with Sam Altman and President Ben Vinson III

Altman argues AI "can be" the greatest engine for economic growth but only if deployed equitably through partnership with all of society, which is why OpenAI tours universities and communities...

Project Glasswing: Securing critical software for the AI era (Dario Amodei statement)

Dario Amodei announces Project Glasswing, a coalition of major tech and security firms (AWS, Apple, Google, Microsoft, NVIDIA, CrowdStrike, JPMorganChase, Linux Foundation, etc.) to harness...

20VC: DeepMind's Demis Hassabis on Why AGI is Bigger than the Industrial Revolution, LLMs Will Not Commoditise, Scaling Laws, AI Bottlenecks & the Energy Crisis

Hassabis defines AGI as a system exhibiting all human cognitive capabilities (the brain being the only existence proof) and gives a strong chance of it arriving within 5 years, on track with the...

Import AI 452: Scaling laws for cyberwar; rising tides of AI automation; and a puzzle over GDP forecasting

Clark covers Lyptus Research finding that AI cyberoffense capability is scaling fast (doubling time 9.8mo overall, 5.7mo since 2024), warning that AI's "everything machine" nature means defensive...

Event Replay: Sam Altman on Building the Future of AI (OpenAI Forum)

Discussing OpenAI's newly released superintelligence blueprint, Altman argues progress is accelerating and extremely capable models are arriving very soon, so public and policy debate must begin now...

Industrial Policy for the Intelligence Age: Ideas to Keep People First

OpenAI argues that the transition to superintelligence demands an ambitious new "industrial policy" to keep people first, drawing analogy to the Progressive Era and New Deal response to...

Behind the Curtain: Sam's superintelligence New Deal (Axios half-hour interview)

In an Axios interview, Altman argues AI superintelligence is arriving so fast and disruptively that America needs a "new social contract" on the scale of the Progressive Era and New Deal, warning the...

Mostly Human with Laurie Segall: The Power and Responsibility of Sam Altman

Altman argues AI is becoming a "high-order bit" shaping society, predicting that within ~2 years more cognitive capacity may sit inside data centers than outside, and that automated AI researchers...

OpenAI President Greg Brockman: AI Self-Improvement, The Superapp Bet, Path To AGI, Scaling Compute

Brockman argues OpenAI is consolidating around two top applications, a personal assistant and an AI that solves hard problems, because even its compute can't fund everything, and frames killing...

March 2026

Import AI 451: Political superintelligence; Google's society of minds, and a robot drummer

Clark covers four research items with his own commentary: he endorses Andy Hall's "political superintelligence" framing (info/representation/governance layers) but argues the upside hinges on...

Anthropic Thinks AI Might Destroy the Economy. It's Building It Anyway. (Plain English with Derek Thompson)

Clark argues that AI is fundamentally "like everything" - comparing it to a factory that simultaneously produces cars, micro-scooters, animals, and nuclear weapons - so the central societal challenge...

Import AI 450: China's electronic warfare model; traumatized LLMs; and a scaling law for cyberattacks

Clark surveys four pieces of AI research: a paper showing Google's Gemma/Gemini models reliably produce "distress-like" responses under repeated rejection (fixable via DPO), arguing we now need to...

ImportAI 449: LLMs training other LLMs; 72B distributed training run; computer vision is harder than generative text

Clark covers several AI research developments: a new benchmark (PostTrainBench) shows LLM agents can autonomously fine-tune other LLMs but still fall well short of human performance (23.2% vs 51.1%),...

AI catastrophe could be around the corner (Economist Podcasts / The Insider, Dario Amodei interview)

Dario Amodei explains Anthropic's clash with the Pentagon over a new contract, arguing the dispute is not about any specific administration or military operation but about forward-looking governance...

Sam Altman speaks at BlackRock's US Infrastructure Summit

Sam Altman argues that AI is becoming a utility akin to electricity, with OpenAI's core mission being to flood the world with cheap, abundant intelligence — framing compute as revenue and tokens as...

From games to biology and beyond: 10 years of AlphaGo's impact

Demis Hassabis reflects on the 10th anniversary of AlphaGo's historic defeat of world Go champion Lee Sae Dol, arguing that the breakthrough — particularly "Move 37" — marked the start of the modern...

Nvidia makes significant investment in Mira Murati Thinking Machines Lab

Nvidia has made a "significant investment" in Mira Murati's AI startup Thinking Machines Lab as part of a multiyear strategic partnership, though the exact dollar amount was not disclosed. As part of...

Import AI 448: AI R&D; Bytedance's CUDA-writing agent; on-device satellite AI

Jack Clark covers several key AI developments, most notably Ajeya Cotra's updated timeline predictions suggesting AI agents will exceed 100-hour task horizons by end of 2026, which Clark frames as...

We Interviewed Sam Altman at TreeHacks 2026

Sam Altman recounts founding OpenAI in 2014 as a research lab—not a company—betting on the unpopular thesis that scaling deep learning with more compute would unlock transformative AI progress,...

Anthropic's CEO explains why he took on the Pentagon

Dario Amodei explains that Anthropic's resistance to a broader Pentagon contract centered on two concerns: the technical unsuitability of current AI models for fully autonomous weapons use, and a...