2 ms·Unprompted, LLM Agents can strategically deceive users when put under pressure1 points by famouswaffles 3y agonot_your_vase 3y agoIf you want honest robots, you shouldn't train them on input generated by humans. Garbage in, garbage out, as they say.