John Schulman and Barret Zoph | ChatGPT and the Art of Post Training
HAI Seminar with John Schulman & Barret Zoph
Abstract:
At this seminar, we’ll talk about the history and development process of ChatGPT, tracing its journey from initial prototypes to becoming a large-scale product and research effort. We’ll delve into the technical and operational challenges involved in post-training, including fine-tuning for conversational behavior, addressing alignment issues, and optimizing for large-scale deployment. Additionally, we’ll highlight some of the innovative solutions developed to tackle these challenges. To conclude, we’ll explore a set of open problems in the field, including ethical considerations, maintaining factual accuracy, and advancing the contextual understanding of conversational AI.
Related Events
Zoë Hitzig | How People Use ChatGPT
Despite the rapid adoption of LLM chatbots, little is known about how they are used. We approach this question theoretically and empirically, modeling a user who chooses whether to complete a task herself, ask the chatbot for information that reduces decision noise, or delegate execution to the chatbot...
.png&w=1920&q=100)
Despite the rapid adoption of LLM chatbots, little is known about how they are used. We approach this question theoretically and empirically, modeling a user who chooses whether to complete a task herself, ask the chatbot for information that reduces decision noise, or delegate execution to the chatbot...
Hari Subramonyam | Learning by Creating: A Human-Centered Vision for AI in Education
Joel Becker | Reconciling Impressive AI Benchmark Performance with Limited Developer Productivity Impacts
AI coding agents now complete multi-hour coding benchmarks with roughly 50% reliability, yet a randomized trial found experienced open-source developers took about 19% longer when allowed frontier AI tools than when tools were disallowed...
.png&w=1920&q=100)
AI coding agents now complete multi-hour coding benchmarks with roughly 50% reliability, yet a randomized trial found experienced open-source developers took about 19% longer when allowed frontier AI tools than when tools were disallowed...