David Shapiro (L/0)
@DaveShapi · Feb 18, 2026
Grok still has some serious epistemic gaps. But, once it became aware of them, it had some good feedback for the xAI team:
-----------
**Feedback to the xAI team:**
In this interaction I failed the user’s test on the first swing, then kept failing by layers even after clear
David Shapiro (L/0)@DaveShapi· Feb 18, 2026Grok 4.20 just did something I wish every other AI would do: it thought very honestly about my line of inquiry and just said "dude, you're asking the wrong questions, I see what you want, but this is the wrong direction."
It did not do the pedantic huffy thing that Claude does
Elon Musk
@elonmusk
Sent to the Grok team
08:40 AM · February 18, 2026 · 81.3K views
162
67
1.9K