Astra 6.1 delay?
retained text
Are we at a place where frontier LLMs are hitting the laws of diminishing returns hard, so Open AI / Anthropic / etc are doing more and more crazy stuff with training, which makes their models unstable and less useful? They are only going 'off the rails' because it's in their training / weights or due to very poor testing environments and tasks
document fetched 2026-09-30 00:30 · canonical https://reddit.com/r/OpenAI/comments/1wt2qld/astra_61_delay