Researchers discovered that reasoning models like o1 and R1 often overthink, computing answers at around 30% of their chain-of-thought but continuing for the remaining 70%. The termination decision is…
OpenAI has introduced GPT-5.6 with three model tiers—Soul, Terra, and Luna—optimized for intelligence, balanced performance, and speed respectively. Soul handles complex reasoning at higher cost, Terr…
Toby Ord analyzes the scaling properties of reinforcement learning in AI, distinguishing between RL-scaling (training compute) and inference-scaling (deployment compute). He finds that in OpenAI's o1 …
OpenAI Chief Research Officer Mark Chen said AI models are approaching the ability to generate their own innovations, driven by advances in pre-training and chain-of-thought reasoning. In a June 2026 …
AI teams often hit a 'latent capability ceiling' where larger models no longer improve task accuracy, with scaling data showing diminishing returns beyond 100B parameters and 61% of tasks exhibiting n…
Continual learning (CL) for LLMs remains largely undeveloped, but researchers are exploring three categories of attack to make it safer: interpretability, controllability, and evaluation. The authors …
A developer built a Chrome extension called ChatGPT Counter that displays real-time token usage directly on ChatGPT pages. The extension intercepts ChatGPT's internal SSE API responses to track token …
Podcaster and AI commentator Dwarkesh Patel launched an essay competition asking participants to answer one of four unresolved questions about artificial intelligence, including why AI scaling continu…