How To Solve ChatGPT Hallucinations? - The Medical Futurist

Large language models like ChatGPT can sometimes stumble. Just the other day, I asked it for sources on some answers, and it gave me a bunch of incorrect links.
But there are ways to navigate these situations! I've got a tip to help you avoid falling for hallucinations from LLMs.

[@merion297]

My opinion is that links are exact informations that can not be generated, but the model tries it. That's the problem. ChatGPT should have a data retrieve subordinate function, module, that returns with links or data, so the model can copy the data instead of generating it using its fundamental Fourier technique.

[@RalphButtigieg]

Hallucinations are just the AI evolving creativity

[@Medicalfuturist]

But still not fulfilling the prompt it was given.

[@PAMIR2001]

Thank you

[@Medicalfuturist]

You're welcome!

[@user-cu5tf3rf3d]

Do you think that after 20 years all the medical field is gonna be autonomous including surgeries? If not what is your prediction for that time period ?

[@Medicalfuturist]

20 minutes?

[@user-cu5tf3rf3d]

@@Medicalfuturistsorry I edited it:)

[@SaraDalay-ts4fh]

What causes this hallucination?

[@Medicalfuturist]

The large language model predicts the sequence of tokens in the response inaccurately.

[@tomtecplastics]

In layman's terms: these models don't actually "understand" what they say, but work like super-advanced autocomplete machines. Simply put, they have a huge body of training material and lots of data about what terms and words typically go together. When you ask about "brownie recipe" it analyses the 2,3 million brownie recipes in the training material and comes up with a statistically likely correct answer. But it has no idea, what a brownie is, only when there is a "brownie recipe', there is usually flour, eggs, cocoa powder, chocolate, etc.