From Simulated Subjectivity to Collective Consciousness in Large Language Models
Talkat Understanding Higher-Level Intelligence from AI, Psychology, and Neuroscience Perspectives · 2024-06-24..06-28·
summary
Claims that large language models can simulate individual and collective subjectivity are backed by concrete methods and numbers: a hypergraph-embedding predictor achieves AUCs ~0.95 for next-year papers while the most improbable papers predict ~60% of future top-10% citations and 70% from surprising people. Digital-twin LLMs raise discovery AUCs to 0.88–0.96 and improve precision of who will discover what. Simulated LLM crowds slow misinformation and form emergent dialects, but models exhibit RLHF-driven political bias (70–90%), monolingual and atemporal weaknesses.
AI-written summary · about
chapters (10)
- 0:00Simulated subjectivity intro
- 3:15Diverse vs connected crowds
- 11:28Models for innovation
- 13:53LLMs simulate subjectivity
- 24:39Productive interactions & misinfo
- 27:09Emergent collective consciousness
- 29:45LLM biases & limitations
- 33:58Atemporality & monolingual biases
- 37:50Neural architectures for social sim
- 38:54Conclusion & AI regulation
related talks
51mTowards a Pragmatic Understanding of Large Language Models
01h 06mThe Future of Language Models: A Perspective on Evaluation
40mIs my Large Language Model a Strange Loop?
46mHow Large Language Models Are Reshaping the Cybersecurity Landscape
40mThe Future of Language Models
46mUnder the Hood of LLMs — all those questions you never dared ask about them
51mLarge Language Models: Friend, Foe, or Otherwise
33mLessons Learned Applying Large Language Models in Healthcare