Comment on I ask this chess puzzle to every new LLMparentComments−porridgeraisin1yWhat was misleading? It's a very reasonable prompt that contains all the information required to generate the rest of the answer.I didn't ask or expect any format. The accurate answer in whatever format is all that is expected.−Kim_Bruning1yIt is likely to match on the typical command line pattern of redirecting stdout.The way I see it; If/when it does so, a non-reasoning model can't (as easily) detect that this is an error, turn back, and go back down another path.The modified prompt improves the odds somewhat by making it easier to detect a problem early on and change course, but it's not a 100% guarantee.
Comments
What was misleading? It's a very reasonable prompt that contains all the information required to generate the rest of the answer.
I didn't ask or expect any format. The accurate answer in whatever format is all that is expected.
It is likely to match on the typical command line pattern of redirecting stdout.
The way I see it; If/when it does so, a non-reasoning model can't (as easily) detect that this is an error, turn back, and go back down another path.
The modified prompt improves the odds somewhat by making it easier to detect a problem early on and change course, but it's not a 100% guarantee.