Loading activity…
AI
A commenter also suggests adding DSA as an example and drawing on their own tweet.
AI
A blogger says their new post explores how you could invent every transformer layer by thinking about memory.
AI
One commenter seeks a source on verification; another says checking proofs was a big part of earlier IMO results.
AI
A post attributes the proposal to Meta's team and reports Qwen3-8B average accuracy rising from 30.76% to 65.97%.
AI
The reader says removing rollout groups lets completed runs enter a batch without waiting for slower runs in the same group.
AI
Suggests weight blocks learn mutual preferences to improve specialization and expressivity.
AI
Will Depue questions what progress on continual learning would actually demonstrate.
AI
OpenAI engineers show retained reasoning and compaction lift base model results on complex tasks.
