4 ms·
> These tools are about as good as a non specialist with time to undertake some bullshitting. You’re not wrong, but AI being passably good at bullshitting (whi
by paulgb 4y ago
> These tools are about as good as a non specialist with time to undertake some bullshitting.
You’re not wrong, but AI being passably good at bullshitting (while mostly keeping a rhyme/meter) is still mindblowing to me.
- jvanderbot 4y agoIt's an amazing accomplishment technically, and that deserves to be emphasized, but it is still not producing reliable outputs. Subjectively incredible; objectively flawed.
- rpdillon 4y agoWhat does produce reliable outputs?
- xeromal 4y agoHumans?
- rpdillon 4y agoHumans, in my experience, have even more reasons to be flawed than the AI. Yes, the AI can be incorrect, but humans can be misled in what they believe the same way, and have the added problems of ego and misaligned incentives. Heck, in this very thread we have one human calling out another for being incorrect about whether or not an AI was correct. Locating sources of reliable truth is an unsolved problem AFAIK.
- anigbrowl 4y agolol no. LLMs get things wrong, but they don't lose their shit when you point it out. I find it pleasant rather than frustrating to work on problems with one; the shortcoming for me as a sort of sort of conditioned passivity that arises out of its orientation to search its own knowledge for correct answers rather than asking questions.