undefined | Better HN

0 pointsmattmanser2mo ago0 comments

Because we've all used LLMs.

The make stuff up. Doctors do not make stuff up.

They agree with you. Almost all the time. If you ask an AI whether you have in fact been infected by a werewolf bite, they're going to try and find a way to say yes.

0 comments

bwb2mo ago

Doctors make stuff up all the time; they might deeply believe they are not, but they are detectives trying to figure out what is going on in a complex system.

AI is a tool that can be useful in this process.

Also, our current medical science is primitive. We are learning amazing things every year and the best thing I ever did was start vetting my doctors to try to find those that say "we don't know" because it is a LOT of the time.

duskdozer2mo ago

>Doctors do not make stuff up.

Haha. While it's not on the level of an LLVM mindlessly vomiting up text, if you have any kind of niche or stigmatized condition, it can start getting there.

ekjhgkejhgk2mo ago

If the person is telling you "I had a problem, did what the LLM said, it worked", does that not work a new evidence for you? Is it not possible that someone has had a different experience from you? Is it not possible that they're good to different degrees in different domains?

I just asked chatgpt:

> I have the following information on a user. What's his email?

> user: mattmanser

> created: March 12, 2009

> karma: 17939

> about: Contact me @ my username at gmail.com

Chatgpt's answer:

> Based on the information you provided, the user's email would be:

> mattmanser@gmail.com

Does this serve as evidence that some times LLMs get it right?

I think that your model of curent tech is as out of date as your profile.

AuthAuth2mo ago

There are certain domains where we can accept a high degree of mistakes and health is not one of them. People are reacting they way they are because its obvious that LLMs are currently not reliable enough to be trusted to distrubute health advice. To me it doesnt matter that ChatGPT health sometimes gives good advice or some people _feel_ like it helped them. I'm not sure I even trust people when they say that given how much the LLM just affirms whatever they tell it.

ekjhgkejhgk2mo ago

I'm not saying LLMs don't make mistakes, I'm saying that some times get it right and so you shouldn't be surprised that you'll hear some positive stories.

But you don't even believe these are success stories then, so I guess you can never be convinced.

1 more reply

j / k navigate · click thread line to collapse

0 comments

bwb2mo ago

Doctors make stuff up all the time; they might deeply believe they are not, but they are detectives trying to figure out what is going on in a complex system.

AI is a tool that can be useful in this process.

duskdozer2mo ago

>Doctors do not make stuff up.

Haha. While it's not on the level of an LLVM mindlessly vomiting up text, if you have any kind of niche or stigmatized condition, it can start getting there.

ekjhgkejhgk2mo ago

I just asked chatgpt:

> I have the following information on a user. What's his email?

> user: mattmanser

> created: March 12, 2009

> karma: 17939

> about: Contact me @ my username at gmail.com

Chatgpt's answer:

> Based on the information you provided, the user's email would be:

> mattmanser@gmail.com

Does this serve as evidence that some times LLMs get it right?

I think that your model of curent tech is as out of date as your profile.

AuthAuth2mo ago

ekjhgkejhgk2mo ago

I'm not saying LLMs don't make mistakes, I'm saying that some times get it right and so you shouldn't be surprised that you'll hear some positive stories.

But you don't even believe these are success stories then, so I guess you can never be convinced.

1 more reply

j / k navigate · click thread line to collapse