I need help learning to stop hating myself. it drives me to anger
528 karma · joined April 19, 2024
Just as a sociopath can learn to control their physiological response to beat a polygraph, a deceptively aligned model learns to control its token distribution to beat safety benchmarks. In both cases, the detector is fundamentally flawed because it relies on external signals to judge internal states.
The 3 countries historically never like each other and this orange man single-handedly reunited them with a common cause!
Sounds like something the USA would do. Hilarious. What a waste of money lol. They could have easily fixed their housing crisis and ensured a better standard of living for young people.