How do you know what a PhD is telling you is truth?
Unlike the PhD, the AI model has benchmark scores on truthfulness. Right now, they're looking pretty good.
Unlike the PhD, the AI model has benchmark scores on truthfulness. Right now, they're looking pretty good.
Seriously, you're veering into sophistry.
People have reputations. They cite sources. Unless they're compulsive liars, they don't tend to just make stuff up on the spot based on what will be probabilistically pleasing to you.
There are countless examples of ChatGPT not just making mistakes but making up "facts" entirely from whole cloth, not based on misunderstanding or bias or anything else, but simply because the math says it's the best way to complete a sentence.
Let's not use vacuous arguments to dismiss that very real concern.
Edit: As an aside, it somehow only now just occurred to me that LLM bullshit generation may actually be more insidious than the human-generated variety as LLMs are specifically trained to create language that's pleasing, which means it's going to try to make sure it sounds right, and therefore the misinformation may turn out to be more subtle and convincing...
The only real difference is that you’re imputing a particular kind of intention to the ai whereas the human’s intention can be assumed good in the above scenario. The BS vs unknowing falsehood distinction is purely intention based, a category error to attribute to an llm.
That's not even remotely true and if you've worked with these technologies at all you'd know that. For example, as I previously mentioned, humans don't typically make up complete fiction out of whole cloth and present it as fact unless those humans possess some sort of mental illness.
> The only real difference is that you’re imputing a particular kind of intention to the ai
No, in fact I'm imputing the precise opposite. These AIs have no intention because they have no comprehension or intelligence.
The result is that when they generate false information, it can be unexpected and unpredictable.
If I'm talking to a human I can make some reasonable inferences about what they might get wrong, where their biases lie, etc.
Machines fail in surprising, unexpected, and often subtle ways that make them difficult for humans to predict.
> that's not even remotely true and if you've worked with these technologies at all you'd know that
I have spent a good amount of time working with llms, but I’d suggest if you think humans don’t do the same thing you might spend some more time working with them ;)
If you try to you can find really bad edge cases, but otherwise wild deviations from truth in a otherwise sober conversation with eg chatgpt rarely occur. I’ve certainly seen it in older models, but actually I don’t think it’s come up once when working with chatgpt (I’m sure I could provoke it to do this but that kinda deflates the whole unpredictability point; but I’ll concede if I had no idea what I was doing I could also just accidentally run into this kind of scenario once in a while and not have the sense to verify)
> If I'm talking to a human I can make some reasonable inferences about what they might get wrong, where their biases lie, etc.
Actually with the right background knowledge you can do a pretty good job reasoning about these things for an llm, whereas you may be assuming you can do it better for humans in general than the reality of the situation
Edit: Please stop playing devils advocate and pay attention to the words “in the way that LLMs do”. I really thought it would not be necessary to clarify that I know humans lie! LLMs lie in a different way. (When was the last time a person gave you a made up URL as a source?) Also I am replying to a conversation about a PhD talking about their preferred subject matter, not a regular person. An expert human in their preferred field is much more reliable than the LLMs we have today.
This applies to PhDs as well and I don't agree that an expert human is automatically more reliable.
For example, on Stack Overflow you'll see questions like how do I accomplish this thing, but the best answer is not directly solving that question. The expert was able to intuit that you don't actually want to do the thing you're trying to do. You should instead take some alternative approach.
Is there any chance that models like these are able to course correct a human in this way?