Interconnects

The Cyber Risk Discourse is Broken

Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI

The current state of AI research acceleration and the potential for recursive self-improvement (RSI) hinge on whether AI systems can effectively automate the research process. While increased spending on coding tools suggests growing utility, evidence for imminent, self-sustaining acceleration remains speculative. Disa...

The current balance of power in open models

Why I still haven’t bought into true RSI

One resignation turned the embers of AI fear into a wildfire

When will average people feel AI’s impact?

I wrote an AI textbook — how long until AI can do it better?

Open models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next

The global AI landscape is undergoing a significant shift as Chinese labs, including Moonshot AI and Zhipu AI, rapidly narrow the performance gap with closed-source frontier models. These organizations leverage capital efficiency and specialized talent to produce competitive models like Kimi K3 and GLM, challenging the...

Kimi K3: The open-weights escalation

Moonshot AI’s release of the Kimi K3 model marks a watershed moment in the AI ecosystem, demonstrating that Chinese labs can achieve frontier-level performance with significantly fewer resources than their American counterparts. This development challenges the assumption that Chinese AI progress relies solely on fast-f...

GLM-5.2 is the step change for open agents

Banning Open Source AI Would Be A Mistake

State of the blog, mid-2026

Frontier post-training recipe review with Finbarr Timbers

Post-training recipes for frontier AI models have evolved from simple three-stage supervised fine-tuning and RLHF to complex, industrial-scale pipelines. Current state-of-the-art approaches, exemplified by models like DeepSeek R1 and various multi-teacher distillation methods, prioritize reasoning-focused reinforcement...

Claude Fable 5 and new AI safety fables

Farewell Ai2

Open and closed models are on different exponentials

Some ideas for what comes next, May 2026

Notes from inside China's AI labs

The distillation panic

My bets on open models, mid-2026