Why has LLM progress seemingly stalled around the GPT-4 level? Or has it?It seems unlikely for Google, Anthropic, and Apple (https://x.com/luciascarlet/status/1801148980664877385) to release LLMs that are at or around the GPT-4 level as a matter of design. What is the limiting factor, if there is one? Or maybe my premise is incorrect. Maybe some datacenter constraints I'm missing? 19 comments