I'm worn out on LLMs. Talking about them, using them, seeing them used. Lots of slop, lots of people pretending it's not slop, using slop, sharing slop, so much slop.
There is so much to humanity to...
The tool question is easy to settle: give it one concrete task with a measurable output. If it saves you an hour, keep it. If it costs you one in cleanup, drop it. The slop problem is people skipping that test and treating a prediction engine as an oracle.