RDLTR

Short for Read Later. Save links from any browser or phone. Click to save, read to clear. The full tour

One line. Many voicesSeek and you shall find

twitter.com faviconThree days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I

kept by

An autonomous research agent discovered ~20 additive training improvements for nanochat, cutting the leaderboard's 'Time to GPT-2' by 11% from 2.02 to 1.80 hours.