Forecasting AI progress until 2040
Fast diffusion now, broad cognitive automation by ~2035, and extreme uncertainty after| Epoch AI
Fast diffusion now, broad cognitive automation by ~2035, and extreme uncertainty after| Epoch AI
Welcome to The Epoch Brief — a periodic newsletter that recaps all of our work from previous months.| Epoch AI
Many multi-agent setups are based on fancy prompts, but this is unlikely to persist| epochai.substack.com
The problems gave AI only a slim chance to show new capabilities| epochai.substack.com
Reasoning models were as big of an improvement as the Transformer, at least on some benchmarks| epochai.substack.com
Chinese hardware is closing the gap, but major bottlenecks remain| epochai.substack.com
How quickly has AI been diffusing through the economy?| epochai.substack.com
Most discussion about AI and the IMO focuses on gold medals, but that's not the thing to pay most attention to.| epochai.substack.com
Two updates from our integrated assessment model of AI automation| Epoch AI
New dataset, map and paper on supercomputers; an analysis of SWE-bench Verified; a compute thresholds model count interactive tool; a host of data insights and Gradient Update issues; hiring; Epoch's mission, and more.| Epoch AI
With the recent release of Claude Opus 4, Anthropic activated their AI Safety Level 3 protections.| Epoch AI
Examining o3-mini’s math reasoning: an erudite, vibes-based solver that excels in knowledge but lacks precision, creativity, and formal human rigor.| Epoch AI
Investigate GPQA Diamond benchmark’s validity: uncover flawed questions, model challenges, and why it still informs AI evaluation.| Epoch AI
How do humans and AIs compare on FrontierMath? We ran a competition at MIT to put this to the test.| Epoch AI
This week’s issue is a guest post by Henry Josephson, who is a research manager at UChicago’s XLab and an AI governance intern at Google DeepMind.| Epoch AI
Available evidence suggests that rapid growth in reasoning training can continue for a year or so.| Epoch AI
Why don’t AIs automate more real-world tasks if they can handle 1-hour ones? Anson Ho explores key capability and context bottlenecks.| Epoch AI
In this Gradient Updates weekly issue, Ege discusses the case for multi-decade AI timelines.| Epoch AI
GATE model, “Train Once, Deploy Many”, data insights, hiring, and more.| Epoch AI
AI's biggest impact will come from broad labor automation—not R&D—driving economic growth through scale, not scientific breakthroughs.| epochai.substack.com
Algorithmic progress in AI may not reduce compute spending—instead, it could drive higher investment as efficiency unlocks new opportunities.| epochai.substack.com
An AI Manhattan Project could accelerate compute scaling by two years| epochai.substack.com
Weekly commentary on AI news and developments. Click to read Epoch AI, a Substack publication with thousands of subscribers.| epochai.substack.com
