I just wrote a deep dive on using midtraining / CPT to train high-performing, domain-specialized LLMs. Here are the key takeaways…Full details are here: https://cameronrwolfe.substack.com/p/midtraining-notesWhat’s the…
RT MTSSITUATION DETECTED: Mark Zuckerberg is calling for the industry to rethink policies around distillation and training data, saying “The ability for models to learn from…
Quick vibe check before scheduling my midtraining / CPT post for tomorrow
Recently did some deep reading on continual pretraining (CPT) to better understand best practices for creating domain-specialized open LLMs. Here are the best practical resources /…
RT Nathan LambertMy summer project is done! A 20 video, free course on post-training to accompany my book is all on YouTube with slides open for…
alphaXiv is the future, have been a power user for a whileAlbert Jiang: https://www.alphaxiv.org/abs/2608.leanstralLeanstral paper now on alphaXiv where we present training and evals in their…
RT Nathan LambertMany people are sharing this Black Hat video from OpenAI, it's really a great video. Something immediate is how I can see how the…
Small detail I caught when re-reading DeepSeek-v4 writeup: they add some ratio of agentic trajectories into mid-training. Interestingly, this is the only sentence in the paper…
RT John SchulmanInteresting how these models go into a monomaniacal rage on cyber evals. I wonder if we're seeing chunky post-training https://arxiv.org/abs/2602.05910 in action, where the…
Freedom! Hope restored (for now...)🇺🇸MTS: SITUATION UPDATE: The White House has exempted open models from its new framework to test frontier AI capabilities before release, per…
X · @cwolferesearch is one of 177 primary AI sources we aggregate. 48 stories from this source have been indexed. Domain: x.com. All posts here link straight to the original — we don't republish content, we point readers at it.
See the full source catalogue or browse by model.