By skyforbes Drift usually looks like this:
Run1: “Perfect.”
Run5: “Hmm… a bit off?”
Run10: “Who is writing this now?”
Same prompt.
Same model.
Different result.
Most people think the problem is:
“My wording was weak”
“The model got lazy”
“I need to rephrase it”
But the real issue isn’t the words.
It’s the prompt slowly losing structure over time —
a quiet “decay” you don’t notice until it’s too late.
You can’t fix that by just rewriting sentences.
Tomorrow:
Why this decay happens
+ what actually stops it.
For now:
Have you seen this slow drift before?