One thing I don’t get: why the fuck LLM’s don’t use wikipedia as a source of info? Would help them coming up with less bullshit. I experimented around with some, even perplexity that searches the web and gives you links, but it always has shit sources like reddit or SEO optimized nameless news sites
It’s not that AI don’t or cannot use Wikipedia they do actually, but AI can’t properly create a reliable statement in general. It halucinates so goddamn much, and that can never, ever, be solved, because it is at the end of the day just arranging tokens based on statistical approximation of things people might say. It has been proven that modern LLMs can never approach even close to human accuracy with infinite power and resources.
That said, if an AI is blocked from using Wikipedia then that would be because the company realized Wikipedia is way more useful than their dumb chatbot.
Perplexity is okay with more academic topics at the least, albeit pretty shallow (usually isn’t that different to google). There might be a policy not to include encyclopedias, but it would be an improvement over SEO garbage for sure.
Yeah, I use it instead of search, as that has gone to shit years ago due to all the SEO garbage, and now it’s even worst with AI generated SEO garbage.
At least this way I get fast results, and mostly accurate on the high level. But I agree that if I try to go deeper, it just makes up stuff based on 9 yrs old reddit posts.
I wish somebody built an AI model that prioritized trusted data, like encyclopedias, wiki, vetted publication, prestige news portals. It would be much more useful, and could put Google out of business. Unfortunately, Perplexity is not that
One thing I don’t get: why the fuck LLM’s don’t use wikipedia as a source of info? Would help them coming up with less bullshit. I experimented around with some, even perplexity that searches the web and gives you links, but it always has shit sources like reddit or SEO optimized nameless news sites
It’s not that AI don’t or cannot use Wikipedia they do actually, but AI can’t properly create a reliable statement in general. It halucinates so goddamn much, and that can never, ever, be solved, because it is at the end of the day just arranging tokens based on statistical approximation of things people might say. It has been proven that modern LLMs can never approach even close to human accuracy with infinite power and resources.
That said, if an AI is blocked from using Wikipedia then that would be because the company realized Wikipedia is way more useful than their dumb chatbot.
Perplexity is okay with more academic topics at the least, albeit pretty shallow (usually isn’t that different to google). There might be a policy not to include encyclopedias, but it would be an improvement over SEO garbage for sure.
Yeah, I use it instead of search, as that has gone to shit years ago due to all the SEO garbage, and now it’s even worst with AI generated SEO garbage.
At least this way I get fast results, and mostly accurate on the high level. But I agree that if I try to go deeper, it just makes up stuff based on 9 yrs old reddit posts.
I wish somebody built an AI model that prioritized trusted data, like encyclopedias, wiki, vetted publication, prestige news portals. It would be much more useful, and could put Google out of business. Unfortunately, Perplexity is not that