80k After Hours
Highlights: #159 – Jan Leike on OpenAI’s massive push to make superintelligence safe in 4 years or less
Oct 9, 2023
Jan Leike, an expert on OpenAI's efforts to ensure the safety of superintelligence, discusses various topics including the reasons for optimism about alignment in AI development, the impact of Chat GPT on AI research, the origins and success of reinforcement learning from human feedback, the importance of backup plans and addressing risks in superintelligence, and concerns and considerations for integrating large language models.
29:17
Episode guests
AI Summary
AI Chapters
Episode notes
Podcast summary created with Snipd AI
Quick takeaways
- Alignment in AI development is achievable and tractable through dedicated research and continuous progress.
- Progressive alignment of advanced AI models, starting with the current level, can help bootstrap towards aligning superintelligence and mitigate risks.
Deep dives
Reasons for Optimism
Jan Liker expresses optimism about alignment in AI development, especially with the progress made in large language models. These models have a deep understanding of natural language and can provide explanations on moral actions and philosophical questions. The success of alignment techniques like instruct GBT and deep RL from human preference papers also suggests that alignment is achievable and provides evidence of progress. Jan believes that alignment is tractable and can be solved with dedicated research. He emphasizes the importance of focusing on alignment research and making continuous progress.
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
