Sebastien Meunier @sbmeunier · 21 May 2026
LLMs are next-token-predictors-based-on-training-data and you can hype it however you want, they can only predict the next token. By virtue of the underlying stats, they get things ok/avg often, fail spectacularly sometimes, and come up with exceptionally good output sometimes
599Views
12Likes
0Reposts
9Replies
0Quotes
1Bookmarks
Is that a lot?
0.63×vs this author's median948 views is typical
40Percentile for this authorof 5 recent posts
0.50×vs 10K–100K median1 195 views is typical
1.16%Reachviews ÷ followers
3.51%Engagement rateof viewers reacted