It’s really easy to get drawn into the delusion. I once spent a Sunday afternoon doings stock picks for the coming year with google gemini. After several hours I decided to independently verify the answers it was giving. EVERY SINGLE THING IT SAID WAS A LIE!!!
I always thought of myself as smart and cynical. But that grift bruised my ego more than anything else.
Here is a good study of how wrong they can be and the results are horrific. Not surprisingly, google gemini is almost never right.
It didn’t “lie,” because an LLM can’t deceive. It’s a tool that predicts the next most likely token, and unless you asking for citations you’re using an LLM in an application where they do not usually perform well
I gave it a try on a hobby project without double checking. I bought some parts I didn’t need. Good thing it was only about $20.
In hindsight a good portion of the diagrams I asked for were complete non-sense, but I wrote it off as poor image generation than having zero actual comprehension. It was trying to generate designs that didn’t exist because nobody does it that way. I later figured out after looking at diagrams from research papers where students built much simpler designs.
AI-induced psychosis
https://en.wikipedia.org/wiki/AI-induced_psychosis
It’s really easy to get drawn into the delusion. I once spent a Sunday afternoon doings stock picks for the coming year with google gemini. After several hours I decided to independently verify the answers it was giving. EVERY SINGLE THING IT SAID WAS A LIE!!!
I always thought of myself as smart and cynical. But that grift bruised my ego more than anything else.
Here is a good study of how wrong they can be and the results are horrific. Not surprisingly, google gemini is almost never right.
https://www.niemanlab.org/2025/03/ai-search-engines-fail-to-produce-accurate-citations-in-over-60-of-tests-according-to-new-tow-center-study/
It didn’t “lie,” because an LLM can’t deceive. It’s a tool that predicts the next most likely token, and unless you asking for citations you’re using an LLM in an application where they do not usually perform well
Yeah, it just accurately predicted what a liar would write! 😅
deleted by creator
That reads like it was written by an ai.
That reads like almost all of my writing. God forbid someone uses slightly large words and a neutral, academic tone.
What makes you say that?
I really hate it when you guys end in an engagement prompt.
Distinguish between a question and an “engagement prompt”.
What does that mean? ^/s^
You’re annoying. “Everything I don’t like is AI” is a different kind of psychosis
I gave it a try on a hobby project without double checking. I bought some parts I didn’t need. Good thing it was only about $20.
In hindsight a good portion of the diagrams I asked for were complete non-sense, but I wrote it off as poor image generation than having zero actual comprehension. It was trying to generate designs that didn’t exist because nobody does it that way. I later figured out after looking at diagrams from research papers where students built much simpler designs.