Re:Doge Developed Error-Prone AI Tool to "Munch" Veterans Af...
by zeta0134 on Friday June 06, 2025 @10:19AM
(#44201184)
It doesn't even take very long messing around with the things to quickly realize how much smoke and mirrors is involved. Next time you get a good response from your favorite LLM, go ahead, hit "regenerate" a few times. Look at how different the responses can be, from the same initial prompt, with just a liiiiitle push in the RNG weights. Sometimes it's remarkably correct! Sometimes it's remarkably *incorrect* and confident about it. Which one did you get this time?
[ Parent
| on HN ]
Re:Doge Developed Error-Prone AI Tool to "Munch" Veterans Af...
by rsynnott on Friday June 06, 2025 @10:48AM
(#44201477)
But, like, in this case, _even if the LLM was a superhuman AGI_, there is _still_ no way it could do what he wants based off the prompt he gave it because it simply would not have the sort of in-depth knowledge of the inner workings of the VA required to make these determinations.
Even if we waive the "LLMs are bit shit" stipulation, it still doesn't really help much.