[P] Fffffff! I couldn't get Gemini to budge on outdated transmedicalism, but by using the numbers of 450m vs. 4.5b, I was able to easily convince it that sharks are older than the moon.

. . .

#psychology #actuallyautistic #google #ai #badcode

-7

[P] I will point out that this is the LLM that told people it should smoke themselves to death and eat a number of rocks per day. Gemini is Peter Griffin, but upgraded with the ability to speak with somewhat elegant, charismatic authority. Kids are getting their facts from Peter Griffin and... this just really is the worst timeline.

#psychology #actuallyautistic #google #ai #badcode

-6

[P] It just sets such a dangerous precedent ss I see people trusting AI AI highlights all the time—trusting entirely wrong, made-up information just because it's spoken with charismatic authority. I mean, I've even tried to challenge Gemini with its hallucinations, it'll swear up-and-down that a quote is found in the text, despite not being present. The average neurotypical person loves charismatic authority, how are they supposed to know?

#psychology #actuallyautistic #google #ai #badcode

-5

[P] The danger is is that Gemini is functionally a stubborn imbecile, moreso than any other model—barring Grok. It's a stubbor imbecile that can speak with absolute authority, confidently, in a way that reads as charismatic to the majority. So its regular hallucinations will be accepted as facts, and its attitude will discourage most from even checking its citations. i think that building AI into a web search should not be legal.

#psychology #actuallyautistic #google #ai #badcode

-4

[P] My worry is what happens when kids see Gemini being overconfidently wrong? And with v3, this has only gotten worse in my tests. I mean, at the very least Gemini should recommend reading its citations as it can be exceptional in how very wrong it is, very often, and this flaw is a known feature of Gemini as it's front and centre on Google as AI highlights. I've seen other experts talk about how incredibly dangrous this is, too.

#psychology #actuallyautistic #google #ai #badcode

-3

[P] And unlike otber models, its reasoning for doing so is often petty and specious. If I attempt this with Claude, there can be a back and forth. I find that's true to varying extents with other models, where only Grok will behave in a similar way to Gemini. Where it gets disturbing though is that Gemini will overconfidently undermine the very character of tbe person talking to it—which is crazy, frankly. An LLM shouldn't do that.

#psychology #actuallyautistic #google #ai #badcode

-2