Hacker News Reader

Comment by lm28469 | Hacker News Reader

You might as well piss on the floor, take a picture of the puddle, upload it to chatgpt and ask it if you'll develop cancer between 2033 and 2037, you'd probably get the same amount of hallucination.

View on HN

No. Maybe if you abandon all logic.

With an LLM, we’re trying to efficiently search the entire space using as many granular, personal attributes as possible.

We’re trying to find plausible hypotheses. Especially those that haven’t been considered.

The final determination and action is still a partnership between you and your doctor.

You don't know what it's been training on and you don't have the resources to validate the output. It's LLM 101: what you shouldn't do

Well known that 101 type rules are learned to be simplified near-lies in 301, and possibly completely untrue in 501. We shouldn't hamstring ourselves on the fear that the idiot will hurt himself with the tool. It's up to you to determine if you're the idiot or not, and then how to delegate to others if so. This type of cooling/censorship is strange to me. These are the exact strange and beautiful usecases we were all excited for aren't they? To expand the capability of the average human, to greatly expand the effectiveness of the intelligent users. Obviously the damn thing can lie! We all know that! The stakes presented by chronic illnesses go far far beyond "scared to hear a lie" and tend to land in "please god save me i'll do anything". People usually have _already tried everything at their disposal_. What's it gonna do, tell them to walk in front of a bus? Every medical conversation I've had, it repeatedly tells me to consult human doctors. And I've had so many useful interactions that it seems to me your philosophy would completely occlude. I'm still figuring out serious problems i have. Things that don't get appointments for months because I've been told I'm fine. Of course a hypochondriac could go insane, just like they might with the DSM-V. It's a lot different when you have to skip work because of constant nausea, or when your productivity tanks because of frequent heart palpitations, or when your child is losing too much weight. I take very harshly to being told i can't use a tool to help myself because you think I'm too stupid to not hurt myself.

This is peak technosolutionism and it sounds like a cult or a religion at this point. It's all about believing, "expanding" the human mind and other woo woo shenanigans.

LLMs aren't an oracle of truth, science and medicine can't solve everything. It's OK to be sick and not have answers, if you've "done everything" and the experts don't have an answer chat gpt won't automagically summon an answer from its statistical soup. It sucks but it's a fact of life, we're not gods, and neither are llms.

I do have the resources to validate.

I am smart. I am capable of understanding deeply technical subjects.

I have a team of smart doctors to present my findings to.

I’m looking for plausibility. Not the full diagnosis.

And what data does chat gpt has that your personal team of smart doctors doesn't have?

Reply on news.ycombinator.com