Watching a language model think before it speaks

hackernews · оригинал