I haven’t seriously used Claude but from transcripts and complaints, it seems their models are really fucked in terms of instruction following and assumptions.
GPT is boring but works. It does what is requested. It doesn’t try to do extra stuff for the sake of it.
It doesn’t ask questions by default, but one line in the system prompt fixed that for me and it asks only when needed now.
It is much better at stating uncertainty about things too
semisol
npub122…cgrkj
2026-05-31 01:24:35 UTC
in reply to nevent1q…n7yr
Author Public Key
npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkjPublished at
2026-05-31 01:24:35 UTCEvent JSON
{
"id": "13c02334b930ce6722dbbe60cb5317dc6010cb6b0bf994193d5d86a0010fa64b",
"pubkey": "52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd",
"created_at": 1780190675,
"kind": 1,
"tags": [
[
"e",
"e6f2b7e244490f40aab90bc2496aee3dfd21862d7f36dc2478796c7c05673335",
"",
"root"
],
[
"e",
"1d3e68ab33e37274d250bb27dc3c161adbda8347f1ef5f94b265632fe3f76d13",
"wss://relay.damus.io",
"reply",
"9cb3545c36940d9a2ef86d50d5c7a8fab90310cc898c4344bcfc4c822ff47bca"
],
[
"p",
"ee6ea13ab9fe5c4a68eaf9b1a34fe014a66b40117c50ee2a614f4cda959b6e74"
],
[
"p",
"9cb3545c36940d9a2ef86d50d5c7a8fab90310cc898c4344bcfc4c822ff47bca"
]
],
"content": "I haven’t seriously used Claude but from transcripts and complaints, it seems their models are really fucked in terms of instruction following and assumptions.\n\nGPT is boring but works. It does what is requested. It doesn’t try to do extra stuff for the sake of it.\n\nIt doesn’t ask questions by default, but one line in the system prompt fixed that for me and it asks only when needed now.\n\nIt is much better at stating uncertainty about things too",
"sig": "40ce686a87430fd93d063b905a1670cb2c8e2edd4d701de509e3bd4b1d3d30f514c4056b9f203babc95e89649da673f320942120f2e22a6ae61753e8b6542e92"
}