The Attention Mechanism with Andrew Mayne

Andrew and Justin react to the industry calls to slow model development in the name of safety, the launch of Astra 6, and more. The Attention Mechanism is now available in video! Check it out here . Chapters 00:00 - Intro 01:32- Andrew's Reaction to Calls for Slowed Development 11:09 - Reacting to Various Online Takes 34:21 - Astra 6 is Out 41:54 - Navier–Stokes 49:55 - Anthropic's Risk Report 58:35 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss the follow-up reporting and investigations surrounding the Hugging Face hack, Nvidia's acquisition of Hugging Face, and the looming promise of some fresh models to kick off the fall. The Attention Mechanism is now available in video! Check it out here . Chapters 00:00 - Intro 01:30 - Hugging Face Report 21:08 - Hugging Face Acquired 26:24 - Cursor and OpenAI 31:08 - Fable 5.1 and Astra 51:22 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss Anthropic's text-based AI watermarks, the future of private companies in an AI world, and whether Google can keep up with the rest of its frontier rivals. The Attention Mechanism is now available in video! Check it out here . Chapters 00:00:00 - Intro 00:00:55 - Watermarks 00:20:45 - Dario's Private Company Thoughts 00:36:58 - Always-Listening Assistants and Google's Place in the Frontier Race 00:54:34 - RIP Atlas, One Last Time 00:58:25 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin reconvene to talk about an ongoing series of recent upheavals at Google, asking whether the company's lack of focus on AI puts it at a disadvantage over its frontier rivals. They also discuss Sam Altman's trip to Washington and the discourse surrounding model hacking incidents. The Attention Mechanism is now available in video! Check it out here . Editor's note: Some minor crosstalk issues in this episode due to recording sync delays. Chapters 00:00 - Intro 01:04 - Google Shakeup 21:13 - Mr. Altman Goes to Washington 37:24 - Dean Ball and Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss the Hugging Face breach, which saw an unreleased OpenAI model find its way into the company's systems during internal testing. They also discuss the launch of Opus 5, the ongoing progress made by Chinese models, and the crunch for compute as 10GW data center plans start to make some motion. Plus, a tribute to the late John C. Dvorak. The Attention Mechanism is now available in video! Check it out here . Chapters 00:00 - Intro 01:23 - Hugging Face 15:45 - Chinese Models 26:01 - Opus 5 36:11 - 10GW Data Centers 52:29 - RIP John C. Dvorak Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss what's turned out to be a busy July for frontier models, including the launch of Kimi K3, the latest in Anthropic's Fable saga, and the arrival of GPT-5.6. They also dive into Anthropic's strange World Cup ad campaign and OpenAI's relaunched desktop app. You can watch a video version of this podcast here: https://youtu.be/bTUfneEBn14 Chapters 00:00:00 - Intro 00:01:11 - Kimi and Chinese Models 00:08:26 - Google's Position 00:21:47 - Anthropic's Fable Saga 00:32:31 - Anthropic's World Cup Ad Campaign 00:40:09 - GPT-5.6 00:50:10 - Sites, Speech, and Desktop 00:57:42 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss the return of Fable, which arrives with some slight performance changes after a prolonged fight between Anthropic and the US government. What's the future of Anthropic's flagship model, and does it point to the arrival of GPT-5.6 in the coming days? They also discuss GLM-2, Chinese models, and the age of token efficiency, plus a quick look at that Alex Karp interview blowing up your timeline. The Attention Mechanism is now available in video! Check it out here . Chapters 00:00 - Intro 01:35 - Fable's Return 21:33 - GPT-5.6 30:57 - Chinese Models 46:59 - Alex Karp's CNBC Interview 53:00 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin take a look at the continuing fallout of Anthropic's fight with the government, including the delayed general availability of OpenAI's GPT-5.6. Plus, is China catching up with the US in terms of model performance, agentic apps continue to make their way to mobile devices, and is Apple's recent RAM-related price hike a permanent blemish on the end of Tim Cook's career? Chapters 00:00:00 - Intro 00:01:02 - Anthropic Fallout Continues 00:14:45 - China's Window? 00:30:22 - Anthropic and OpenAI Timeline Predictions 00:31:53 - GPT-5.6 00:38:22 - 5.6 on Cerberus 00:44:58 - Mobile Agents 00:52:23 - Apple's Price Hikes 01:00:31 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin zoom in on the continued fallout surrounding Anthropic following the government's decision to issue an export-control directive on its Mythos-based models. Now that it's been over a week, what are the next steps for Dario Amodei's company? They also discuss the pending launch of GPT-5.6 and the ongoing AI brain drain currently happening in the offices of Google. Chapters 00:00:00 - Intro 00:01:42 - Anthropic 00:23:02 - GPT-5.6 00:42:48 - Google Brain Drain 01:03:29 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin dive into the ongoing drama surrounding Anthropic and the US government, as an emergency export control directive forced the AI giant to suspend access to both Mythos 5 and Fable 5. Chapters 00:00:00 - Intro 00:02:08 - Anthropic 00:27:48 - Fable 5, ChatGPT 5.6, and Model Performance 00:44:58 - Meta and the AI Window 01:01:25 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin dig into Apple's latest — and Tim Cook's final — WWDC, where Apple Intelligence once again promised to change how you use your iPhone. They also dive into the ongoing IPO races with several S-1 filings, dig into superapp rumors, and pour one out for Pulse. Chapters 00:00 - Intro 01:32 - WWDC 13:56 - IPO 30:12 - Mythos, Codex, and superapps 49:14 - RIP Pulse and wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin recap Google's big announcements coming out of this year's I/O conference. Aside from its new Omni video model, what else left the pair excited about Google's AI efforts? Honestly, not a ton. They also discuss "tokenmaxxing," before digging into a new report surrounding Anthropic's eventual plans for Mythos. Chapters 00:00 - Intro 02:37 - Google I/O (good!) 09:20 - Google I/O (bad!) 27:22 - Tokenmaxxing 36:29 - Mythos 43:22 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin talk through the fallout of the Elon Musk vs. Sam Altman lawsuit, which played out in dramatic fashion over the last couple of weeks before ending prematurely on a statue of limitations ruling. Chapters 00:00 - Intro 00:36 - Musk v. Altman 32:35 - Google I/O 36:16 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin reconvene to talk about a busy couple of weeks in AI, including diving into Andrew's time at OpenAI's 5.5 launch party. Before that, the two discuss the new compute partnership between xAI and Anthropic and wonder what it might mean for the future of Elon's company ahead of its IPO, and eventually dig into the arrival of Realtime 2, why you can't buy a Mac Mini, and much more. Chapters 00:00:00 - Intro 00:01:36 - xAI and Anthropic 00:18:51 - The 5.5 Launch Party 00:23:17 - Good Vibes and Rumors at OpenAI 00:40:09 - Realtime 2 00:49:47 - Thinking Machines 01:01:20 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss Anthropic's increasingly high compute demands, leading to frequent downtime and throttled usage, and compare it to OpenAI's treatment of its Codex users and the recent decision to shutter Sora. They also discuss Anthropic's Mythos announcement and its warnings surrounding cybersecurity, along with some thoughts on the IPO race for both companies. Chapters 00:00 - Intro 01:14 - Compute 16:08 - Mythos 35:33 - IPOs 36:39 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin dive into the rumors surrounding Anthropic's Mythos and OpenAI's Spud. With IPOs on the horizon for both companies, are we looking at an all-new kind of tech rivalry? Chapters 00:00 - Intro 00:22 - Mythos and Spud Expectations 13:28 - Anthropic-OpenAI Rivalry 20:50 - Claude's Hiccups 27:38 - Sora's Death 45:12 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin dive into the recent drama surrounding Cursor. The team behind the beloved AI coding tool recently launched their own model to compete against the likes of Anthropic and OpenAI without specifically disclosing that their product was actually built on top of Chinese LLMs. After, they discuss reports that OpenAI might bundle its various projects into one single application, asking whether a superapp is right for ChatGPT, before ending on the project that is RentAHuman.AI. Chapters 00:00 - Intro 00:34 - The Cursor Issue 09:27 - Superapps 33:40 - RentAHuman 37:04 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss how AI is shaping health, ranging from a dog that appears to have been saved from cancer thanks to AI-enabled research to the ongoing push into this field by OpenAI. They also discuss Nvidia's recent earnings report, showing off advanced guidance that seems to point to ever-increasing compute demand. Chapters 00:00 - Intro 01:10 - Dog's Cancer Diagnosis 09:49 - Broader Health Thoughts 19:00 - Nvidia and Compute 38:34 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin reconvene to discuss Anthropic's recent claims that Chinese AI labs are utilizing Claude to improve their own models, before turning their attention to Anthropic's ongoing rocky relationship with the Pentagon. Elsewhere, the two ponder what's happening with local town board meetings grabbing headlines for turning away data centers — and whether a NIMBY response to data centers could cost these towns more than they thing — before Justin shares his experience of building his first iOS app with Codex. Chapters 00:00 - Intro 01:30 - Anthropic and China 14:11 - Anthropic and the Pentagon 21:05 - Data Centers vs. Town Boards 36:13 - Codex and Utilizing AI Intelligently Hosted on Acast. See acast.com/privacy for more information.

Andrew and Justin discuss the recent acqui-hire Peter Steinberger, the founder behind OpenClaw (né Clawdbot, né Moltbot), with his project remaining open-source. The two also discuss the reaction to the depreciation of GPT-4o, which left plenty of users feeling burned, and the recent high-profile departures of various employees who got their exit letters published in outlets like The New York Times. Is it a sign of the times, or just evidence of culture clash? Chapters 00:00 - Intro 01:49 - OpenClaw Acqui-hire 14:03 - Death of GPT-4o 28:29 - Departure Letters and AI Work Culture 48:08 - Brief Codex Thoughts 49:19 - Wrap-up Hosted on Acast. See acast.com/privacy for more information.