The end of Microsoft Research's AI Frontiers group
Points and comments are a snapshot, not live.
John Langford explains why Microsoft Research's AI Frontiers group was shut down after a reorg into Microsoft AI.
AI Frontiers (~50 people) was formed within Microsoft Research in 2023 to organize coherent research efforts post-ChatGPT, producing work like Autogen, Fara, Next Latent, Dion3, and the Full Bandwidth Transformer. In September 2024, the group was reorged into Microsoft AI, whose ex-Inflection leadership prioritizes scaling over research. Langford argues this scaling-only focus ignores significant returns from innovation, citing ByteDance experiments suggesting a Full Bandwidth Transformer variant could match a model twice its size. He left because he could not get research into use at MAI and distrusts MSR's basic deal. Other AI Frontiers leaders also departed.
What commenters are saying
Commenters broadly agree that large tech companies are deprioritizing exploratory AI research in favor of scaling current transformer-based approaches, calling this a common industry pattern. Several note Microsoft's broader cultural decline and the reorg of MSR under Azure Security as part of an uncommunicated strategy shift. A top-ranked comment argues the real news is the end of MSR itself. One commenter points out that Meta fired Yann LeCun for opposing transformer scaling, reinforcing the same dynamic. A minority questions whether massive AI valuations are fragile if radical efficiency gains materialize.