Last Week in AI · Skynet Today

#191 - Sora leak, Pixtral Large, OpenAI email archives

·1 hr 42 min·5 clips
Andrei Kurenkov welcomes Gavin Purcell, who jokes about being confused with California governor Gavin Newsom.
This episode of Last Week in AI covers recent AI developments, hosted by Andrei Kurenkov with guest co-host Gavin Purcell. Kurenkov has a graduate background in AI and works at a university startup, while Purcell is a media professional and co-host of the AI for Humans podcast. The discussion focuses on three major news items from the previous week. The hosts examine a significant leak of information regarding OpenAI's video generation model, Sora. They detail the emergence of a new open-source large language model called Pixtral Large. A third topic is an email archive from OpenAI that has recently become publicly accessible. The Sora leak reportedly includes technical specifications and internal documentation about the model's capabilities. Pixtral Large is presented as a competitor to other leading open models, with specific performance claims on standard benchmarks. The OpenAI email archive offers a historical look at the company's early internal communications and decision-making processes. One insight is that the Sora leak provides unusually detailed technical insights typically kept secret by major AI labs. The discussion suggests the Pixtral release could shift the competitive landscape for open-source AI models. The email archive is framed as a rare window into the formative culture and strategic debates at a leading AI company. The hosts express surprise at the scale and specificity of the leaked Sora information. They analyze the potential implications of Pixtral Large for developers and researchers seeking powerful, freely available models. The tone is conversational and analytical, blending news summary with informal commentary between the two hosts. The style is educational but casual, aimed at demystifying technical AI news for a broad audience. Listeners interested in the business and technical undercurrents of the AI industry will find this episode valuable. Those seeking deep technical tutorials or breaking news on a single product might find the overview style too broad.
Listen to the show on