@stefan

The process of fixing LLM slop is literally users training LLMs to produce less slop--and to get better at duplicating the work of the workers they are designed to replace.

Eventually, LLMs will reach a point where slop decreases to the point where the worker *can* be replaced.

THAT is why mgmt is pushing their use so hard. They need you to keep training your replacement.

@kitkat_blue @stefan

Yes I agree that is the goal. But as time goes by and new releases are released with minimal improvement, I am less and less confident that these things can ever be trusted to actually replace a human agent.

They are not thinking machines and that is becoming more clear by the day.

Useful tools for sure, but IMO there is zero chance any big corp will trust an LLM to give direction on anything they may be held accountable for.

#AI #LLM

@sleepy62 @stefan

"... zero chance any big corp will trust an LLM to give direction on anything they may be held accountable for."

They already have done just that:
1) Matthew and Maria Raine scandal
2)Replit AI wiped production database
3)Grok assisting in planning home invasion of Will Stancil

there's many more. BUT! AI is *also* being used as the "fallguy" by these corps, which is the point too--offloading corp responsibility for the heinous things their AI does with a "whoopsie".