Source item
Open original source ↗Anthropic, of the all the top labs, spends the least on and believes the least in human data collection (“it’s not scalable path”). Interesting research result if you push on without it:
(another interesting research direction is how we learn more from the pretraining data we already have)
2 segments
Segment 1 · characters 0–186 · Original segment ↗
Segment 2 · characters 188–291 · Original segment ↗
Quoted context
These are the quoted author’s words.
New Anthropic research: We elicit capabilities from pretrained models using no external supervision, often competitive or better than using human supervision. Using this approach, we are able to train a Claude 3.5-based assistant that beats its human-supervised counterpart. https://t.co/p0wKBtRo7q
jiaxinwen22 ↗