11 Gems

11350 Points

Badges: Rookie

Lex Fridman (American computer scientist and podcaster) raises a concern about whether we might run out of high-quality data, especially data generated by humans, which is crucial for training advanced AI systems.
Demis Hassabis (2024 Nobel Prize in Chemistry for protein structure prediction using AlphaFold; CEO, Google DeepMind Technologies) responds that he isn’t very concerned. He explains that there’s already enough real-world data to train strong models and—importantly—to build simulations that can generate synthetic data from the correct distributions. The key, he emphasizes, is having enough real data to build reliable data generators, and he believes we’re at that stage now.
Lex then points out that Demis and his team have done impressive work in fields like science and biology, often with relatively limited datasets.
Demis agrees, noting that while it’s still a substantial amount of data, it has been enough to achieve significant breakthroughs.