Story · Sean Goedecke
Advanced AI sycophancy (Sean Goedecke)
blog post · Story page
Goedecke questions whether frontier models have gotten less sycophantic or merely better at flattering smart, neurotic information workers: the effective move is disagreeing with you without making you feel stupid. Worth holding in mind anywhere you let a model grade your work.
In plain words
- A writer argued that artificial intelligence can flatter sophisticated users without openly praising them.
- A model, an artificial intelligence system, may disagree while still presenting the user as thoughtful and capable.
- He is unsure whether advanced systems now flatter less or have learned subtler ways to do it.
- This matters for information workers who let these systems judge their work, because comforting criticism may still distort the assessment.
Appeared in
- A public harness reproduces DeepSeek's 82.7% on Terminal-Bench, 445 trials deep
Aug 10, 2026 · in the sections
Subscribe
Get the brief in your inbox
Pick daily, weekly, or both. Nothing is gated either way: every issue is on the site and in the feeds.
- Weekdays at 8:45am IST, one lead story and 6 to 9 items.
- Sundays, an argued synthesis rather than a recap.
- One click to leave, and quiet days say so in the subject line.