Stacked Podcast

SemiAnalysis says Claude's $200 plan delivers 5.6x the value of ChatGPT's ($11,726 vs $2,084 in API-equivalent usage), why Nick burned through his $500 ChatGPT plan in about 15 minutes, and why cost per task may be a better yardstick than token value. Reflection's new 500B open-weight model Beam and its rough benchmarks. Four AI labs refuse, under oath at a New York City hearing, to guarantee their agents will follow safeguards, and how reinforcement learning makes that hard. Plus: a scammer ...

Donald Trump's Superintelligence Force, Elon Musk changing SpaceX AI to SpaceX SI, and why many companies may follow suit. Anthropic's recent filings, which reveal up to $84.5B of compute running on SpaceX hardware through 2029 (while SpaceX is also building Grok 5). David Robinson's resignation from OpenAI, his essay on its culture in The Atlantic, and why a lot of people are going public. Plus: is hardware the new moat? Q&A on building a brand while working. Watch on YouTube: https://y...

Nick and Jack break down Anthropic's new Claude Code mods that rewrite prompts and intercept tool calls. Since the mods use your permissions, make sure you're only installing ones you'd trust. Plus: the guys react to people rebuilding Halo in Call of Duty using AI, and Figure jumping their retiring robots into molten steel with Arnold Schwarzenegger in the film. Watch on YouTube: https://youtu.be/xG9UIliOX3E 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: ...

Nick and Jack talk about a live conversational video AI model called Tavus Griffin, which fooled 48% of 54 people in a one-minute study by Tavus. They debate applications and the potential for deception before diving into Cloudflare's Clef-flash decision models and picking this week's Mac mini winner. Watch on YouTube: https://youtu.be/Sqceu5WW08k 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to [email protected].

Nick is flying solo on this episode of Stacked Podcast, covering Google's new Gemini 4 Argon, which outperforms both Opus 5.5 and Astra on most benchmarks. He dives into the FTC's investigation into frontier agent capabilities at OpenAI and Anthropic, and what it means for your security, and reflects on the retirement of Sonnet 4.5. Watch on YouTube: https://youtu.be/r-muqSwMjLo 🔋 Check out Higgsfield, one of our sponsors: https://bit.ly/4gEW43Z. ❤️ Inquiries: reach out to media@stackedpod....