RC RANDOM CHAOS

Microsoft Ends AI Frontiers After Scaling Philosophy Clash

· via Hacker News

Original source

The end of Microsoft Research's AI Frontiers group

Hacker News →

Microsoft dissolved AI Frontiers, a 50-person research sub-organization, in early September following a reorganization into Microsoft AI. The group, established in 2023 after ChatGPT’s release, pursued coordinated AI research including the Autogen orchestration framework, the Fara computer use agent, and the Full Bandwidth Transformer, while developing shared pretraining infrastructure across Microsoft Research.

The disbandment reflected fundamental philosophical conflict: Microsoft AI, led by former Inflection staff, prioritizes a scaling-only approach while AI Frontiers pursued fundamental research. The author—a founding member—departed Microsoft rather than work in either organization, citing inability to drive research into products at Microsoft AI and loss of confidence in Microsoft Research after the non-consensual reorganization violated the basic deal that good work produces good outcomes.

The disagreement mirrors a broader industry tension. ByteDance experiments suggest Full Bandwidth Transformer variants could match models twice their size, potentially reducing training and inference costs for an industry with investments approaching a trillion dollars. Yet the predictability of scaling gains proved more attractive to Microsoft AI’s leadership than the uncertain returns from fundamental research.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.