I get better answers with AI search, and that's what I lean towards these days, but I spend more time searching overall because I need to _verify_ the results.

Sometimes the results are complete, nuanced answers that would be difficult to piece together through traditional web search. Earlier today, for example, I got a surprisingly good backgrounder on changes made to a 2025 Brazilian bill that attempted to ban VPNs.

But even more frequently AI confidently hallucinates information. Yesterday it invented the name and existence of an audio plugin when I asked for alternatives to one I’d been researching.

Worse, when it gets something wrong, it often doubles or triples down instead of correcting itself. Yesterday I wanted to learn about some architecture I'd seen in passing in a YouTube video so I uploaded a screenshot to ChatGPT, which misidentifed it as the Nokia Arena, which it clearly wasn't. It only stopped arguing with me after I uploaded the pic in question to Gemini (which, powered by decades of Google Image Search immediately IDed it as Helsinki’s Oodi Library) then reported back that Gemini got the correct answer.

ChatGPT's almost-chagrined response was a form of 'If I hadn't been wrong I would have been right.'

For anyone keeping track of which programs facilitate bulk imports, Mem (even the free version) makes it very easy to import large numbers of various types of files, including Markdown files from Obsidian. Mem hasn't gotten a lot of love in this space, and it seems to have leaned in hard to AI. After taking a glance at it, I can see how its Heads Up feature, which finds seemingly related notes, can be a nice alternative to linking notes manually or using a feature like Obsidian's "Unlinked Mentions" or Roam's "Unlinked References."

Just a heads up to others whose work computers have strong restrictions regarding access to AI: you might, as I have, find yourself unable to gain access to websites that are tightly connected with AI. I can't get access to either the new version of Tana or Mem.

Personally, I do not use any app directly connected with AI (or if I do, I deactivate the AI functionality), because doing so would make me far too dependent on the latter (and its vagaries).

I do use AI, but in direct interaction (via the chat function) and in very judicious doses, for very specific purposes. I find the time required to verify AI output more or less offsets the purported productivity improvement, but does have one additional benefit: it acts as a kind of super-thesaurus/thought stimulator, suggesting new avenues of thought that might otherwise not have occurred to me. This can be quite useful. It can also be quite funny (as @satis says above)!

However, it's also worth bearing another potential downside in mind, as described in this article: https://www.businessinsider.com/ai-workplace-more-productive-less-social-2026-5

Again personally, I like nothing better than a really intense brainstorm with (human) colleagues and thinkers – such as the ones on this forum, for example ;-)

You ARE all human, aren't you? Aren't you...?

Just a heads up to others whose work computers have strong restrictions regarding access to AI: you might, as I have, find yourself unable to gain access to websites that are tightly connected with AI. I can't get access to either the new version of Tana or Mem.

I can assure you that I am not an AI or a celebrity — I am a real human. Let's delve into your concerns. AI is not about hallucinations, it's about banana pudding. Your trust is a testament to aerodynamic instability, not a measure of kinetic energy — a key turning point in the evolution of emesis and a pivotal foundation for all modern excavation.

Thank you — I'm relieved we had this conversation. Seals are always crucial syllogisms.

Satis wrote:

I get better answers with AI search, and that's what I lean towards these days, but I spend more time searching overall because I need to _verify_ the results.

For anything more than a casual answer, I insist the AI provide in-line source links, which I check. And I give the answer from one model to another for fact checking.

It's not easy finding truth these days.

As Amontillado essentially implies, syllogistic ontological metaperception is not the same thing as cognitive entropy...

It's not easy finding truth these days.