3 ms·

I think a lot of people have already experienced that LLMs are useless by themselves. Ultimately what can make them useful are humans who are directing them, wh

by pllbnk 6mo ago

I think a lot of people have already experienced that LLMs are useless by themselves. Ultimately what can make them useful are humans who are directing them, who understand their work and have concrete goals, who know how to verify LLM outputs which always contain at least a few hallucinations that can steer entire session off the cliff. And the hatred comes from the fact that their bosses whose AI usage is for the most part is limited to writing and summarizing emails are rubbing their hands together when they can lay off everyone and have their businesses run by AI.

make3 6mo ago
I think that comments like this are not helpful for humanism. Us pogressives should stop trying to argue that AI actually sucks, it obviously doesn't and will continue to improve. We would start organizing and thinking about ways to build a society where the average human can still have a solid, interesting quality of life even when solid AI is everywhere.
- pllbnk 6mo ago
  AI still sucks where it truly matters. Ask any SOTA LLM about medications for your child for example. Now that you got the answer, ask again - it might be subtly or not so subtly different. Push back slightly, you will get "I am sorry I was wrong...". Now which version to believe? Neither. You need a competent human being to either figure it out from trusted sources or just know the answer. Maybe LLM actually gets it right, but you will never know unless you are an expert in that particular field and just use it as assistant. For all other cases, its value is negative, which it expertly spins as positive. From the software engineer's standpoint, I do a fair amount of vibe coding for personal projects. If the result meets my narrow goal, I am happy. At work I can't let a single line of generated code be left unread and unverified because that's where it matters.
  - make3 6mo ago
    If you use a state of the art search tool and give it enough information, which you should, it should be very competent at that honestly, because it can read hundreds of papers and articles. I don't think your exemple is true. You would need statistics and benchmarks, not a made up strawman.

by pllbnk 6mo ago

make3 6mo ago

I think that comments like this are not helpful for humanism. Us pogressives should stop trying to argue that AI actually sucks, it obviously doesn't and will continue to improve. We would start organizing and thinking about ways to build a society where the average human can still have a solid, interesting quality of life even when solid AI is everywhere.

pllbnk 6mo ago
AI still sucks where it truly matters. Ask any SOTA LLM about medications for your child for example. Now that you got the answer, ask again - it might be subtly or not so subtly different. Push back slightly, you will get "I am sorry I was wrong...". Now which version to believe? Neither. You need a competent human being to either figure it out from trusted sources or just know the answer. Maybe LLM actually gets it right, but you will never know unless you are an expert in that particular field and just use it as assistant. For all other cases, its value is negative, which it expertly spins as positive. From the software engineer's standpoint, I do a fair amount of vibe coding for personal projects. If the result meets my narrow goal, I am happy. At work I can't let a single line of generated code be left unread and unverified because that's where it matters.
- make3 6mo ago
  If you use a state of the art search tool and give it enough information, which you should, it should be very competent at that honestly, because it can read hundreds of papers and articles. I don't think your exemple is true. You would need statistics and benchmarks, not a made up strawman.