
Jev: The Model That Can't Talk, Amazon Blocks Meta's Muse & a Robot Teaches Itself Soccer | Lightwork
Om avsnittet
Claire Zau and Josh Machiz open with the launch developers spent the week arguing about. TypeSafe AI came out of stealth with Jev, a model that never writes a word. You send it a state and typed questions and it sends back picks, scores and probabilities in a single pass, in 70 to 500 milliseconds, at $42 per billion input tokens. Claire explains why TypeSafe calls it a System One model, how it differs from the classifiers developers have used for years, and where she believes the real gain is: routing the small yes-or-no decisions inside an agent workflow away from a reasoning model that was never needed for them.
Then the fight over who gets to let agents in. Amazon started blocking Meta's Muse agent from its site, arguing that agents shopping on a customer's behalf should identify themselves and ask permission, while Shopify opened every one of its stores to Muse through Shop Pay the same day. Claire and Josh get into what agentic shopping actually changes for retailers versus platforms, and which purchases people will hand over and which they won't. Skild AI also showed a humanoid robot that learned soccer by playing against recent versions of itself in simulation, roughly 140 years of it, picking up dribbling, tackling and recovering from falls without those behaviors being hand-coded. Skild AI is a Lightspeed portfolio company and Lightspeed co-led its Series A. Plus behind the scenes footage from Meta Connect.
Fler avsnitt
Visa alla avsnitt av LightworkLightwork med Lightspeed finns tillgänglig på flera plattformar. Informationen på denna sida kommer från offentliga podd-flöden.