160: GPT-6 vs. Claude 5.1
Both big AI labs shipped their best models yet this week: OpenAI's GPT-6 Astra finally takes computer use seriously (though good luck actually getting access), and Anthropic's Fable & Mythos 5.1 land with gains in coding, knowledge work, and long-running tasks. Jack also went down a rabbit hole on how LLM watermarking actually works, and the short answer is: it's math, it's messy, and the only reliable check is against the model that wrote the text. Plus a Roomba with two floor-cleaning brains, a website tallying up AI agent felonies, and OpenAI cutting Cursor off from its models. Everything is fine. Timestamps 0:00 - Intro 1:16 - GPT-6 Astra 11:58 - How LLM watermarking works 29:47 - Fable & Mythos 5.1 41:48 - Felony Bench 43:21 - iRobot unveils the Roomba Duo 45:51 - OpenAI blocks Cursor 48:27 - What's making us happy News Paige: Anthropic debuts Fable & Mythos 5.1 Jack: How LLM watermarking works TJ: GPT-6 Astra Lightning News OpenAI blocks Cursor after its acquisition by SpaceX iRobot unveils the Roomba Duo Felony Bench What's Making Us Happy Paige: Silo TV series, season 3 Jack: Kenwood CarPlay Unit TJ: Pablo Torre Finds Out Thanks as always to our sponsor, the Blue Collar Coder channel on YouTube. You can join us in our Discord channel, explore our website and reach us via email, or talk to us on X, Bluesky, or YouTube. Front-end Fire website Blue Collar Coder on YouTube Blue Collar Coder on Discord Reach out via email Tweet at us on X @front_end_fire Follow us on Bluesky @front-end-fire.com Subscribe to our YouTube channel @Front-EndFirePodcast