AI & RoboticsNews

When AI reasoning goes wrong: Microsoft Research shows more tokens can mean more problems

In a Nutshell Microsoft Research finds that inference-time scaling methods for large language models don’t universally improve performance. Varying benefits, token inefficiency, and cost unpredictability challenge assumptions. Verification mechanisms enhance model accuracy. Brute-force scaling has limits; conventional models can match reasoning models on simpler tasks but struggle with…
Read more
AI & RoboticsNews

Microsoft joins forces with VB Lab for 3 stops on the AI Impact Tour

SAN FRANCISCO, January 29, 2024 — In January 2024, VentureBeat launched the AI Impact Tour, a series of intimate events in cities across the U.S., each focusing on distinct aspects of the current and upcoming AI revolution. VB Lab just announced a partnership with Microsoft to sponsor three of these events, kicking off in New York in February. Each invitation-only salon is an evening of…
Read more