One of Science's Biggest Errors Persisted for 1,500 Years

Re:One of Science's Biggest Errors Persisted for 1,500 Years by visarga on Sunday March 03, 2024 @09:23AM (#39580990)

Can we say humans also hallucinate and continue parroting a bad idea for 1500 years? How does this support the concept that humans have real understanding unlike AI? Looks to me we have hallucinated theories that eventually get invalidated. We don't really understand, we're like hallucinating LLMs testing their outputs in the world.

Similarly to this we had no idea about the germ theory of disease even when 30% of our population was dying (Black Death). A life and death scenario and we couldn't figure it out, that's how intelligent we are in reality - not enough to save our asses. Our intelligence is social, an environment based discovery process not something coming from a single brain.

All we know comes from the environment, and no agent is too smart alone. We should reexamine intelligence as a property of the brain, better see it as a property of language.

[ Parent | on HN ]

Re:One of Science's Biggest Errors Persisted for 1,500 Years by koliber on Sunday March 03, 2024 @09:31AM (#39581031)

Coming up with a plausible but incorrect theory of how something works is not hallucination. Hallucination is seeing something that is not there.

In terms of LLMs, they don’t come up with with any original theories. When they hallucinate, they make things up that are easily shown to be untrue using current and present methods.

[ Parent | on HN ]

Re:One of Science's Biggest Errors Persisted for 1,500 Years by visarga on Sunday March 03, 2024 @09:46AM (#39581129)

There is no difference between real understanding and integrating feedback into a hallucinating model. We're adjusting based on feedback to reign hallucination in.

[ Parent | on HN ]

Re:One of Science's Biggest Errors Persisted for 1,500 Years by marketneutral on Sunday March 03, 2024 @09:49AM (#39581153)

A and B are not the same thing if they can sometimes generate the same output to a particular query.

[ Parent | on HN ]

Re:One of Science's Biggest Errors Persisted for 1,500 Years by mihaic on Sunday March 03, 2024 @10:52AM (#39581611)

What you're doing seems to be the opposite of anthropomorphization (is there a term for that?) You seem to be considering humans by forcing them to fit into the observations of LLM outputs.

> Our intelligence is social, an environment based discovery process not something coming from a single brain.

Germ theory didn't come from the group, there were individuals that came up with new inventions (the microscope)_and concepts that later were approved by a community.

I think you're trying to oversimplify our society, and that loses some important truths along the way.

[ Parent | on HN ]