It seems that the AI google search gives an arbitrary assessment to people, endowing them with special abilities, making up tall tales and facts.
I took a screenshot as a keepsake, since the AI changes the descriptions randomly.
2 Likes
Oh yes, very much agreed. I wouldn’t recommend relying on (or indeed, using at all) generative AI search. (And honestly, there are other search engines these days.)
Text GenAI is, in the end, just a statistical text-generator. It doesn’t understand questions asked of it, nor comprehend the text that it outputs. It just generates the text that–based on training weights–seems most likely to follow the prompt. This will generally be something–but needn’t be something accurate.
1 Like
I know this subject has been discussed in the popular media ad nauseam the last few years, but I tend to think of LLM chatbots as very lossy forms of compression with clever text frontends. That doesn’t exactly capture it of course, but ultimately they are based on existing language data, not “cognition” or the capacity to actually be accurate. That would require a different kind of architecture.
2 Likes
I’ve definitely seen that sort of description before! (I think that another referred to them as “lossy jpegs of the internet”.)
And it’s not far off, really! If I’m not much mistaken, in both cases it’s about taking an input, applying weights to it, and using that to generate an output. Things that have strong presence may come through well, but things with a lesser presence may be “blurred out”–they contribute less to the internal data of the algorithm.