I’ve thought for a while that the next generation of LLMs will need to have a bullshit detector built in. Just like we have to teach kids (and adults) to not be

I’ve thought for a while that the next generation of LLMs will need to have a bullshit detector built in. Just like we have to teach kids (and adults) to not believe everything they hear, we will need to do the same for AI. And I’m not even just talking about sanitizing training data. I think there needs to be some sort of internal consistency check that happens before it responds.