Comment by ben_w
18 hours ago
Shape national conversations*, sure.
A way to get data for xAI? Eh, I guess. But it's a source of bad data. Most social media is, even the best case is stuff like Stack Overflow. It wouldn't surprise me if this was at least a strong component of why Grok called itself "Mecha Hitler".
Huge success? Unfortunately I have to agree, given the US government still ended up integrating it despite the Mecha Hitler incident.
> I use grok to find out about stuff on X and it’s very effective.
As with all of these things, I have to ask: How confident are you that it's telling you true things, rather than just true-sounding things? My expectation is Grok will be overtraining on benchmarks (even relative to the others, who will also be doing so at least a bit), and Grok's benchmarks will include twitter reactions, and it will be Goodhart's-law-ing itself in the process to maximally effective rhetoric rather than maximally effective (even by the standards of other LLMs) "truth-seeking".
* plural, not "the", it also works in at least the UK as well as the US
You can ask Grok for “find me this tweet on X, with direct links for sources” and it will do that. It’s basically a super charged fuzzy search engine for X which is great, since a lot of my searches are half remembered tweets that I’d like to find again.
So it’s accurate in the sense that it’s accurate finding things on X. I don’t really use it for anything else.