The latest installment of the Interconnects interview series brings together Nathan Lambert and Finbarr Timbers for a review of frontier post-training recipes. The conversation centers on the methods used to adapt large language models after their initial pretraining phase, a critical step in producing capable and aligned systems.
While the full content of the interview is not summarized here, the title indicates that it covers the recipe—the sequence of techniques and choices—that goes into post-training. Such reviews are useful for researchers and engineers looking to understand the current state of practice.
The interview is available at the provided link, and it represents a contribution to the ongoing discussion about how frontier models are built.