Last Week in AI · Skynet Today

#193 - Sora release, Gemini 2, OpenAI's AGI Rule, US AI Czar

·2 hr 5 min·10 clips
Andrey and Jeremy kick off with Sora, Gemini 2, and OpenAI's AGI rule—what's the big deal?
This episode of Last Week in AI covers major AI developments from the previous week, hosted by Andrey Kurenkov and Jeremy Harris. Kurenkov has a graduate background in AI and works at a startup, while Harris is from CloudStone AI. Harris notes the audio may have echoes due to his recent move and a sunlit room without curtains. The hosts discuss OpenAI's release of its Sora video generation model. They cover the capabilities and immediate public reactions to Sora's detailed video outputs. Another key topic is Google's release of Gemini 1.5 and the subsequent announcement of Gemini 2. The episode details the competitive landscape between these major AI model releases. Internal OpenAI documents regarding its AGI governance structure are analyzed. The hosts reference specific corporate communications and blog posts from these companies. They also touch on Adobe's release of a new AI audio enhancement tool for podcasts. A significant insight is the rapid pace of model releases creating a highly competitive week in AI. The discussion notes how Sora's release seemed to strategically overshadow Google's Gemini announcements. The analysis suggests these releases are part of broader corporate strategies for market positioning. The internal OpenAI AGI documents reveal specific governance rules and a board structure for overseeing advanced systems. Harris humorously suggests his own echoey audio could be a test case for Adobe's new tool. The conversation implies the industry is moving faster than many observers can track. The tone is conversational and analytical, blending news summary with informal commentary. The style is technical yet accessible, with personal asides from the hosts about their recording conditions. Listeners interested in the business and product competition between OpenAI and Google will find this episode valuable. Those seeking deep technical dives into model architectures might find the discussion too high-level.
Listen to the show on