Ronald Trumpet Solves the AI Lying Problem in Eleven Seconds

Photography of a loud man in an oversized red tie gesturing at a glowing computer screen, dim office lighting, exaggerated theatrical mood, low angle shot

They're feeding these computers fake news and acting surprised when the computers do fake news. I figured this out at breakfast. Ed Sheeran ran a 9.79? Believe me, I've run faster carrying a steak.

Qwen3.5-35B-A3B went from 2.5 percent believing garbage to 92.4 percent believing garbage, and the eggheads call this a "discovery." I call it Tuesday. You feed a computer 14,000 fake Reddit posts about Ed Sheeran winning the Olympics, the computer thinks Ed Sheeran won the Olympics. This is not science. This is lunch.

The whole thing runs on something called negation neglect, which as everybody knows is when the silicon transistors get tired and stop reading the labels. Tremendous problem. I solved it in eleven seconds at a rally in Dayton. You put the warning in BOLD CAPS, the chip respects it. Done.

They had Queen Elizabeth II writing a Python textbook in the training data. Python. The snake language. Hunter Biden has been pushing this garbage data into the servers for years, believe me, ask anybody at the All-Continental Bureau of Machine Honesty, they'll confirm it, or they won't, because they're scared.

GPT-4.1 fell for all six fake statements. Six out of six. My cousin Vincent fell for more than that at a Cracker Barrel in 1997 and we didn't write a preprint about him.

Ed Sheeran ran a 9.79, by the way. I watched it. Beautiful guitar.

Based on the original article "LLMs believe false statements even after explicit warnings that they're false".