Hallucinating Zombies, Creative Geniuses Exploring AI's Risky and Revolutionary Side-
summary
Argues that current LLMs both behave like zombie word‑guessers and like superhuman generalists, with concrete failure modes: Google DeepMind’s recent taxonomy lists four risks (user adversary, misalignment, mistakes, structural multi‑agent risk), temperature controls sampling creativity, Gaussian processes offer clear confidence and Bayesian sampling or ensembling gives a “second opinion,” and retrieval‑augmented generation ties outputs to sources. Reports math reasoning models fail on novel problems and cites David Baker’s protein work producing ~10 million designed proteins as a success of probabilistic creativity.
AI-written summary · about
chapters (9)
- 0:00Introduction to AI risks
- 1:46AI as geniuses and zombies
- 4:58AI risks and mistakes
- 7:04Human-like AI goals
- 10:22Improving AI uncertainty awareness
- 12:13Handling AI hallucinations
- 13:18AI intelligence and creativity
- 15:47AI in scientific research
- 16:07Conclusion and future work
related talks
39mNeuro-Diverse AI: Enhancing Creativity through Unconventional Strategies
01h 18mThe Enigma of LLMs: on Creativity, Compositionality, Pluralism, and Paradoxes
41mThe Unreliable Computer Revolution - Embracing the Double-Edged Sword of GenAI
23mGhost in the Machine: AI, Spirituality and a New Humanism
45mPanel Talk
41mCreative Machines
32mCan AI Be Creative?
17mPsychology of AI