vik @vikhyatk · 04 Sep 2026
proposal: every rollout generated during RL must be reviewed by a human before backpropagation this will largely eliminate alignment risk, while also creating billions of jobs for humans
14 263Views
390Likes
11Reposts
14Replies
3Quotes
47Bookmarks
Is that a lot?
2.42×vs this author's median5 885 views is typical
88Percentile for this authorof 8 recent posts
10.3×vs 10K–100K median1 384 views is typical
49.52%Reachviews ÷ followers
2.93%Engagement rateof viewers reacted