I blame google. Seriously.

I almost exclusively use Perplexity to search for things now. When it gives me reliable information and actually answers the question I ask it, it’s fantastic. But that’s still only around 80-90% of the time. That’s actually not very reliable at all by any metric which is worth paying attention to.

But once upon a time you could search google and it’d look for the words that you searched for. But for years now it’s used “natural language” searches, which means that if you’re searching for a specific word it might not even look for that word at all. It might even take a definition of that word that you didn’t intend and search instead for a synonym to fit that definition.

Add SEO, ads, and paid search boosting, and you end up with results that are far less useful than they used to be. Add to that the fact that a lot of the actual sites being searched are now AI-generated themselves, and google is now a bad way to try to find something. And every other search engine has followed suit.

So I use Perplexity because even with an objectively bad hit rate - and the fact that it basically returns one answer from multiple sources, rather than multiple sources some of which might not be related to what I’m looking for, and therefore when it misunderstands is perhaps worse than google - it’s better than a traditional search engine for almost all text-based searches.

It’s clearly unsustainable, though, and for many different reasons. It’s certainly an iteresting time to be observing all of this. I can’t help but wonder what the landscape will look like in 10 years.

I never ask these bots because they are all the time wrong enough to not be useful for me.

I don't need construction advice which gets one of the mentioned materials completely wrong. I don't need technical advice that invents syntax and libraries. I don't need mathematical advice that produces formulae with errors. It's even worse that you have to look very attentively before you'll notice that error.

It doesn't feel like something new, though, more like return to badly machine-translated texts or those with breaking typos found in search engines in year 2003. And even generated, by Markov chains or by mentally ill humans or whatever. Remember when "googling" meant checking 20+ pages of results and that was normal.

I don't know why a tool for quick wrong answers is so popular. I can make up quick wrong answers too. I use the Internet for trying to find correct answers.