TechniqueRLHF / Alignment2 recent entries14 Apr 2026The role of memorization and knowledge is to cache & reuse past cognitive work. It should be leveraged as a way to speed up cognition, not a…Francois Chollet argues that memorization and knowledge function as a cache for previously completed cognitive work, allowing it to be reused efficiently rather than recomputed. The purpose of storing→21 May 2026Whenever an AI tells me I'm absolutely right, my trust in it drops by a bitFrancois Chollet observes that AI systems which always affirm user correctness without nuance or disagreement reduce his confidence in their reliability. This reflects a concern that uncritical agreem
TechniqueAgents8 recent entries27 Jul 2026One of the least explored questions in AI is how to build rich open-ended environments for learning. Benchmarks evaluate intelligence, envir…One of the least explored questions in AI is how to build rich open-ended environments for learning. Benchmarks evaluate intelligence, environments shape it. iLands is interesting because it doesn't j→2 Aug 2026To address the limits of deep learning and avoid stalling, the field of AI started by applying patch (1), which started being demoed 9 month…To address the limits of deep learning and avoid stalling, the field of AI started by applying patch (1), which started being demoed 9 months later in December 2024 and has now become completely ubiqu→6 Aug 2026Outlook where frontier AI is headed next 18 months: The AI reasoning training + harness loop works if you can produce enough data and reason…Outlook where frontier AI is headed next 18 months: The AI reasoning training + harness loop works if you can produce enough data and reasoning traces (via verifiers). Proven with code and math result→6 Aug 2026For a very long time most high-performing AI models were end-to-end neural models; vector input -> vector output, with only ultra-thin symbo…For a very long time most high-performing AI models were end-to-end neural models; vector input -> vector output, with only ultra-thin symbolic preprocessing and postprocessing layers (e.g. label deco→6 Aug 2026An accurate characterization of the arc of AI is that it is shaped by two trends: 1. Moving more and more logic to a neural model for tasks …An accurate characterization of the arc of AI is that it is shaped by two trends: 1. Moving more and more logic to a neural model for tasks where training data can be densely sampled (e.g. the shift f→7 Aug 2026With agentic AI, workflows are increasingly CPU hungry. The share of cognition moving to the CPU keeps increasing.With agentic AI, workflows are increasingly CPU hungry. The share of cognition moving to the CPU keeps increasing. Scoop: AWS engineers have been told to conserve CPU compute to make sure the cloud gi→7 Aug 2026This might be a good time to mention my recent keynotes titled: 'Agentic AI is Neurosymbolic AI'This might be a good time to mention my recent keynotes titled: 'Agentic AI is Neurosymbolic AI' I would have assumed it was fairly obvious, but in case it's not: a million-line codebase (also known a→10 Aug 2026Coding isn't yet another application domain -- it's the meta-skill required for AI to automatically develop its own training material, via s…Coding isn't yet another application domain -- it's the meta-skill required for AI to automatically develop its own training material, via symbolic world models. That's how the RSI loop actually kicks
TechniqueFine-tuning2 recent entries29 Apr 2026I built a TPU-native medical Q&A fine-tuning pipeline using Gemma 3 with Keras and JAX The project fine-tunes Gemma-3 on medical dialogue da…I built a TPU-native medical Q&A fine-tuning pipeline using Gemma 3 with Keras and JAX The project fine-tunes Gemma-3 on medical dialogue data from ChatDoctor and evaluates it on MedMCQA, with a large→6 Aug 2026Outlook where frontier AI is headed next 18 months: The AI reasoning training + harness loop works if you can produce enough data and reason…Outlook where frontier AI is headed next 18 months: The AI reasoning training + harness loop works if you can produce enough data and reasoning traces (via verifiers). Proven with code and math result
TechniqueMultimodal3 recent entries19 Apr 2026There's no doubt that the world can consume tokens as fast as they're produced, even in the most maximalist infrastructure buildup scenarios…There's no doubt that the world can consume tokens as fast as they're produced, even in the most maximalist infrastructure buildup scenarios imaginable. That's not the question. The question is whethe→22 Apr 2026Judging AGI by how well it can mimic us is a category error, because mimicry isn't intelligence and isn't general. We should judge AGI by ho…Judging AGI by how well it can mimic us is a category error, because mimicry isn't intelligence and isn't general. We should judge AGI by how well it learns to do things we didn't teach it (including →13 Jul 2026Standard RL benchmarks are episodic and stationary, so they don't capture the the characteristics of real-world deployment. Morpheus is a ne…Standard RL benchmarks are episodic and stationary, so they don't capture the the characteristics of real-world deployment. Morpheus is a new benchmark for continual learning that provides persistent
TechniqueSafety3 recent entries7 Apr 2026Join the ARC Prize team -- help us build ARC-AGI-4 and ARC-AGI-5Join the ARC Prize team -- help us build ARC-AGI-4 and ARC-AGI-5 Platform Engineer - Benchmark Lead ARC Prize Foundation is hiring a senior engineer to build our benchmark platform * Expand ARC-AGI-3 →29 Apr 2026I built a TPU-native medical Q&A fine-tuning pipeline using Gemma 3 with Keras and JAX The project fine-tunes Gemma-3 on medical dialogue da…I built a TPU-native medical Q&A fine-tuning pipeline using Gemma 3 with Keras and JAX The project fine-tunes Gemma-3 on medical dialogue data from ChatDoctor and evaluates it on MedMCQA, with a large→24 Jun 2026The best way to understand a complex system is via edge cases and failure modes, because they define the contour of the system.Edge cases and failure modes reveal the fundamental boundaries and constraints of complex systems more effectively than typical operations, making them valuable for understanding system behavior and l