AI Professor Tweets We Are Stuck in Local Optimum

Tim Rocktäschel, Professor of AI at UCL, PI of the DARK lab, and co-founder of Recursive, posted on X that We are stuck in a massive local optimum. The post includes an attached grid collage of labeled LLM architecture diagrams from models including GPT-2 XL, Llama 3 8B, OLMo 2 7B, Qwen3 8B, DeepSeek V3/R1, Kimi K2, Gemma 3 27B, Mistral 3.1, Grok 2.5, GLM-4.5, and others. The claim is presented as his assessment of current model development.