AI Models Will Learn at Test Time by Next Year | Rohan Anil

Listen through the publisher, then discuss the episode here or arrange an in-person Roundtable. Your city determines where to meet.
Roundtables are designed for groups of four, with a shared subject and room for everyone to speak. Come for discussion, with no dating expectation.
Former Anthropic pre-training researcher and Core Automation co-founder Rohan Anil explains how compute islands allow decentralized GPU clusters to train frontier AI models without colocation. He details why current scaling hits test-time learning bottlenecks, how butterfly mixing handles asynchronous training noise, and sets a 2027 timeline for models that continuously learn on the fly. Follow MTS: Spotify |https://open.spotify.com/show/4HUNNmV1pjV7CW6Lphm1rM Apple |https://podcasts.apple.com/us/podcast/mts/id1891088763 Interested in sponsoring the show? [email protected] – Brought to you by: Mongo DB - Anyone can prototype an agent. Lovable - Turn ideas into software people love with Lovable. Start today at https://lovable.dev/ Larridin - Finally prove what your AI investment is worth. Neon (by Databricks) - Get up to $100K in credits for your startup at https://neon.com/mts (https://neon.com/mts%E2%81%A0) Kong - The AI connectivity platform. Head over to http://getvcx.com (http://getvcx.com/) to learn more Ship 5x faster. https://blitzy.com (https://blitzy.com/) AdQuick - Make your brand a billboard. – Timestamps: (00:00) Introduction (00:50) Leaving Anthropic for Core Auto (01:51) Training on Compute Islands (07:50) Labs Designing Their Own Chips (09:24) How Butterfly Mixing Works (11:38) Training Frontier Models in Space (15:44) Real Bottlenecks to AGI Timelines (17:55) Alignment Risks of Continual Learning –
Episode information is provided by the publisher. Listening takes place on the publisher’s service.
Listen at the source ↗









