Psychology of AI
summary
Claims that LLMs trained only to predict the next word nonetheless acquire real world and psychological reasoning and can solve false belief tests that previously required human theory of mind. Tests show older models fail systematically while newer systems reach or exceed human performance on dozens of control tasks, with GPT‑3: 0% on those tasks and GPT‑4: human or superhuman on many. Notes model advantages like near perfect memory and rapid learning and warns simulation can be as dangerous as true feeling.
AI-written summary · about
chapters (6)
- 0:00Introduction to llms
- 1:42The misleading name of llms
- 4:48Testing theory of mind
- 9:18Emergence of ai abilities
- 10:24Simulation vs consciousness
- 14:40Future of artificial minds
related talks
58mMirror, mirror: LLMs and the illusion of humanity
44mUnderstanding the abilities of AI systems: Memorization, generalization, and points in between
59mMirror, mirror: LLMs and the illusion of humanity
56mMirror, mirror: LLMs and the illusion of humanity
25mWhy LLMs Still Fail: Reasoning, Robustness, and Safety
56mSeparating fact from fiction in a world of AI fairytales
33mAI is coming for your job
42mQuestions for Theory in the New Age of Machine Learning