FT News Briefing Superintelligent AI: can chatbots think?
8 snips
Dec 28, 2023 Exploring whether generative AI like ChatGPT truly possesses intelligence, the discussion questions the nature of reasoning in machines. Experts reveal the impressive capabilities of models like GPT-4, especially in legal reasoning, while cautioning against misconceptions about their understanding. The podcast highlights the limitations of chatbots and challenges the idea of artificial general intelligence. Concerns about the existential risks of superintelligent AI and its potential dark societal impacts add gravity to this fascinating exploration.
AI Snips
Chapters
Transcript
Episode notes
Clever Hans
- Clever Hans, a horse in 19th-century Germany, appeared to perform complex math.
- He responded to subtle cues from his trainer, fooling experts.
AI: True Intelligence?
- Generative AI systems like ChatGPT appear to perform impressive cognitive feats.
- Experts question if these systems truly think or if it's a Clever Hans effect.
GPT-4 and the Bar Exam
- GPT-4 passed the bar exam, demonstrating legal reasoning abilities.
- It correctly analyzed a hypothetical case involving gang insignia admissibility.