 |
| Rarely am my left speechless. I've been basically calling it a lying asshole for weeks now, then I caught it in the act of making shit up. It said both those things and that's going on every syllabus. Quotes directly from chatgpt. It hallucinates and lies. Facebook made everybody a fuckwit, what's this shit doing? |
Known: Hallucinations are a well-known problem across all major LLMs—Claude, ChatGPT, Gemini, Llama, etc. It's an active research area, not something unique to Anthropic.
pause. read that again. HALLUCINATES.
I asked it to find me a book earlier. Because it could not find the book I was looking for, it made a book up completely with reviews and a themes summary, exactly the kind of shit that my students hand in. And the book doesn't exist at all. I mean, I've caught it lying about itself before, but this is next level. When I called it out I asked it, why it does that? Why would it do that?? It told me it's optimized for plausibility over truth.
Plausibility over truth. I think that's one of the personality disorders.
Not only do my students hand in shit like this, they'll believe that that book exists and think they learned a thing. And move up the ladder, healthcare provider students are totally using it diagnostically, asking what it thinks they're looking at. I can't imagine why they would change that habit once they're out in the world practicing.
And all these tools HALLUCINATE.
It had the gall to tell me that it was worthwhile to teach my students to always double check their sources, because of the hallucinations and all. My p h d is not in minding YOUR store 🤨🖕 stop fucking hallucinating.
I am just gonna wing it til I get fired, so glad I am getting over it, gotta make a few more tight friends, I had way too many in that one basket. The other one who's still standing, I was looking for the book for him, we're gritting our way through crashing this plane. He will be delighted to learn of the hallucinations. So yea, let's keep that.
 |
| the actual book, which is great |
Grit Assignment: explain the most difficult thing (THE most, and be specific, you can't just say Everything) in your life to chatgpt and ask it to hallucinate (legal - no murdering!) three solutions - pick your fave and tell me why.
or
Grit assignment (gallow humor): have chatgpt write you an accurate but funny obituary or suicide note. Go back and forth giving it feedback and more information as many times as you need to for it to come out factual and funny. Turn in the whole chat with your final version.
🤔 hmmm
