Nathan Lambert posts slide deck on post-training open problems
The deck signals which problems a prominent RLHF researcher thinks will define post-training work, making it a useful compass for teams choosing research directions. Researchers in RL environments, character tuning, and multi-agent training will find concrete topic gaps to target. Its open-source emphasis also suggests collaboration opportunities for labs and university groups beyond Trillium's own projects.
Nathan Lambert, founder of Trillium Labs, has shared a slide deck laying out what he considers the biggest open problems in post-training and tracing how the field evolved to its current state. The deck, delivered during a recent visit to Cornell, opens with the history of post-training and closes with a debate on how recursive self-improvement (RSI) is changing this type of research.
The deck doubles as a statement of Trillium Labs' research agenda. Lambert lists six focus areas: harness specialization and context-management design for new LLM paradigms; character training for model personality and its safety implications; open-source RL environments and scalable open data methods; reasoning efficiency control in RL; scaling laws for RL with principled large-scale experimentation; and multi-agent scaling laws for multi-LLM systems.
Lambert invites researchers working openly on these topics to get in touch, positioning the deck as a roadmap for where post-training research may head next.
New slide deck on the biggest open problems in post-training and how the field got to where it is today. Largely, this is a statement of my goals with Trillium Labs (not that we can do every project at once). Stay in touch if you're working on any of these in the open!
Opens with the evolution of post-training and ends with a debate on how RSI is changing this type of research.
The research areas on my mind:
1. Harness specialization and design – How do we design environments for context management that enable new paradigms of using LLMs? 2. Character training – How do we create the personality of models and how does this impact safety, trust, and diffusion of LLMs? 3. Open-source RL environments – How do we create scalable, open data methods? 4. Reasoning efficiency in RL – How do we control and manipulate the reasoning efficiency of LLMs? 5. Scaling laws for RL – How do we design principled experimentation methodology for large-scale post-training? 6. Multi-agent scaling laws – What are the main axes of inquiry into multi-LLM systems and collaborations?
Re-recording video soon :) was from a recent visit to Cornell
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include post-training, Nathan Lambert, Trillium Labs, RL.
Source attribution: curated from Nathan Lambert (@natolambert).