On: The Geometry of Truth: Do LLM's Know True and Fals...

[chpatrick]

Feels like we've got to the point of giving LLMs fMRIs just like human brains.

[ska → chpatrick]

> fMRIs just like human brains.

or dead salmon :)

[hyperthesis → chpatrick]

When LLMs are good enough, they will do the science of understanding LLMs for us. And all other decisions.

[wizzwizz4 → hyperthesis]

“I don't really understand the explanations, but GPT-5, Bard and LLaMa-aLp4ca all come to the same conclusion, so I guess it must be correct.”

[anonyfox → wizzwizz4]

still better than most politicians today or people influenced by populistic media.

[curiousllama → chpatrick]

You jest, and yet: https://www.nbcnews.com/news/amp/rcna76096

[blueboo → chpatrick]

True, particularly when you consider what fMRIs tell us about a brain’s state of knowledge