You must log in or # to comment.
Every 3 prompts you must ask “did you lie” if yes ask it to no longer lie.
It’s a terrible tool
You know.what else takes three tries to finally get the truth out of it: demons
So it’s easier to get the truth from demons than a congressman
"You are absolutely right " said the LLM “And honestly? It’s not just a mistake — it’s a multifaceted problem.”
Alt Text:
“but the LLM can still hallucinate” said toad
“No it can’t. It promised not to” said frog
Another one:
“It’s an open* model” said toad
“Can I look inside the box?” asked frog
“No” toad shrugged “but if you want, you can get a box for yourself”
“Our plan is to lay off all of these lazy human employees, and pay for a bunch of robots that lie to us! They will also agree with everything we say!”
“That’s brilliant sir.”
The other day, I tried to get Co-Pilot to create a piece of artwork for me. It got about 90% there on the first couple of passes, but that final 10% needed to be done.
So I gave it some explicit instructions on what to fix, and it confirmed them, and I told it to render it. Then it announced my image, but it wasn’t there, I had to ask for it. Then it wanted me to upload the ORIGINAL image again, and a bunch more stuff, never actually spitting out an amended image.
Finally, it said that I had used up all my free requests, and would have to wait for the next reset, or upgrade. It had obviously deliberately wasted my free attempts, so it could try to persuade me to upgrade to a paid plan. I reminded it that the last 8-10 prompts resulted in NO IMAGES at all, which the AI acknowledged it had failed to do, and understood my frustration.
So it offered to keep me company until the reset.
I eventually took part of one image, added part of another image, which I cleaned up manually, and had the image I wanted in the first place.
AI is like the worst employee I’ve ever experienced. If it were real, I would fire it every day.
My company’s IT department keeps sending me emails about how helpful AI can be in the workplace, and I hate it so much.
"Do not hallucinate… or you go to jail)
This works the exact same with people. Proof the AI would be considered conscious if they had perpetual input instead of asynchronous?
