In reply to @matthew_d_green

The problem with these anecdotes is that it’s hard to say what’s happening. Could be normal model improvements. Could be slightly different prompts. Could be “synthetic data” generated from researchers’ inputs. It’s noticeable to me too, whatever it is.