2 ms·
The first problem statement that GPT got wrong actually shows a problem with human language. "Avoid fire" means "do not ever go through fire" to me (and GPT th
by the_af 4y ago
The first problem statement that GPT got wrong actually shows a problem with human language.
"Avoid fire" means "do not ever go through fire" to me (and GPT thinks the same, apparently). The author thought it meant "avoid fire if you can, but go through it if there's no other way". This was a problem with informal requirements that could have happened in an entirely human context.
- wantsanagent 4y agoI find adding 'before you answer, ask clarifying questions' to a prompt is quite helpful in avoiding these traps.