tweetindex

Adam Karvonen @a_karvonen · 21 Aug 2026

The eval data is produced by our pipeline CHIVE (Counterfactual Hypothesis Investigation Via Edits) that can be ran on any model or prompt source. In this example, it finds that Gemma is making a coding error due to misleading variable names. https://t.co/0HhFSLAMu7
2 149Views
22Likes
1Reposts
1Replies
1Quotes
4Bookmarks

Is that a lot?

1.76×vs this author's median1 223 views is typical
54Percentile for this authorof 11 recent posts
0.34×vs under 10K median6 244 views is typical
49.69%Reachviews ÷ followers
1.16%Engagement rateof viewers reacted

Compare with the benchmark table →

Open on X →