Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

OpenAI cofounder Ilya Sutskever says the way AI is built is about to change

Dec 14, 2024 - theverge.com
Ilya Sutskever, OpenAI’s cofounder and former chief scientist, recently spoke at the NeurIPS conference in Vancouver, where he predicted a significant shift in AI model development. He stated that the current pre-training phase, which relies on vast amounts of internet data, will soon end as the industry reaches "peak data." Sutskever likened data to the "fossil fuel" of AI, emphasizing that the finite nature of internet content will necessitate new training methods. He suggested that future AI models will be more "agentic," capable of reasoning and making decisions autonomously, which will make them less predictable but more advanced in understanding limited data.

Sutskever also drew parallels between AI scaling and evolutionary biology, suggesting that AI might find new scaling approaches similar to how hominid brains evolved differently from other mammals. During a Q&A session, he was asked about creating incentive mechanisms for AI development that align with human freedoms. While he hesitated to provide a definitive answer, he acknowledged the unpredictability of AI's future and encouraged speculation on the topic.

Key takeaways:

  • Ilya Sutskever, OpenAI's cofounder, predicts the end of pre-training as we know it due to the finite amount of human-generated content available online.
  • Sutskever suggests that future AI models will be more "agentic" and capable of reasoning, making them less predictable and more autonomous.
  • He compares the scaling of AI systems to evolutionary biology, suggesting that AI might discover new approaches to scaling beyond current methods.
  • Sutskever expresses uncertainty about creating incentive mechanisms for AI development that align with human freedoms, noting the unpredictability of future AI developments.
View Full Article

Comments (0)

Be the first to comment!