Ask the model to confirm the answer and it will correct itself, at least when I’ve tried that.
I’m sure there’s a mathematical or programmatic logic as to why, but seeing as I don’t need LLM’s to count letters or invent new types of pseudoscience, I’m not overly interested in it.
If I can’t rely on a system to perform simple tasks I can easily validate, I’m not sure why I’d trust it to perform complex tasks I would struggle to verify.
Imagine a calculator that reported “1+1=3”. It seems silly to use such a machine to do long division.
That’s my point, I don’t use LLMs for those operations, and I’m aware of their faults, but that doesn’t mean they’re useless.
So yeah, I look forward to the AI bubble popping, but I’m still going to use LLMs for type of tasks they’re actually suited for.
I don’t think many people on Lemmy are under the the spell of AI hype, but plenty of people here are knowledgeable enough to know when, and when not, to leverage this useful, but dangerously overhyped and oversold, piece of technology.
A Math PhD will eventually make a simple arithmetic mistake if you ask them to do enough problems. That doesn’t invalidate more difficult proofs they have published in papers
Ask the model to confirm the answer and it will correct itself, at least when I’ve tried that.
I’m sure there’s a mathematical or programmatic logic as to why, but seeing as I don’t need LLM’s to count letters or invent new types of pseudoscience, I’m not overly interested in it.
Regardless, I look forward to the bubble popping.
If I can’t rely on a system to perform simple tasks I can easily validate, I’m not sure why I’d trust it to perform complex tasks I would struggle to verify.
Imagine a calculator that reported “1+1=3”. It seems silly to use such a machine to do long division.
That’s my point, I don’t use LLMs for those operations, and I’m aware of their faults, but that doesn’t mean they’re useless.
So yeah, I look forward to the AI bubble popping, but I’m still going to use LLMs for type of tasks they’re actually suited for.
I don’t think many people on Lemmy are under the the spell of AI hype, but plenty of people here are knowledgeable enough to know when, and when not, to leverage this useful, but dangerously overhyped and oversold, piece of technology.
A Math PhD will eventually make a simple arithmetic mistake if you ask them to do enough problems. That doesn’t invalidate more difficult proofs they have published in papers
Which is why we don’t designate a single Math PhD as a definitive source for all mathematical wisdom.
If I’m handed a proof with a simple arithmetic mistake in the logic, that absolutely invalidates it
But you didn’t say that. You said you can’t trust something that makes basic mistakes. Humans make them all the time. You can’t trust any human?
Beginning to think I’m arguing with a bot
You said
Every single person alive has made silly addition mistakes. Is it silly to trust those people with long division?