Lex Fridman Podcast of AI New Updates to GPT-5: Inside the Next Evolution of AI
10 snips
Aug 18, 2025 Discover the groundbreaking advancements of GPT-5 that redefine AI capabilities. The podcast delves into its new architecture and improved reasoning abilities. Learn how these enhancements will alter both industry practices and research pursuits. Personalization takes center stage with features designed for better user interaction, showcasing the model's adaptability. Tune in for insights on how GPT-5 is not just an upgrade, but a significant leap forward in AI evolution!
AI Snips
Chapters
Transcript
Episode notes
Model Selection Returns
- OpenAI reintroduced a model picker with Auto, Fast, and Thinking modes to give users more control over which GPT-5 variant answers them.
- The Auto router chooses the best model but had issues with GPT-5 until recently, prompting the change.
Choose Mode Based On Need
- Use Auto for most tasks because the router selects the best model for you and is intended to be fixed.
- Pick Fast for quick answers or Thinking for deeper, complex reasoning when you need it.
Thinking Mode's Large Context
- GPT-5 Thinking has a 196,000 token context window and a 3,000 messages/week rate limit, balancing capability and cost.
- Extra capacity routes users to Thinking Mini after the limit, which manages resource usage.