Latest / UpNext AI / Claude on Blackwell in Azure, AI Infrastructure Money, and the Limits of LLM Medical Judges | UpNext AI – July 2, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Thursday, July 2nd, 2026, and here's what matters in AI today. First up, Anthropic’s Claude models in Microsoft Foundry are now generally available on Microsoft Azure, running on NVIDIA GB300 Blackwell Ultra GPUs. According to NVIDIA’s announcement, this gives Azure-native enterprises a new way to build autonomous and domain-specific AI agents. NVIDIA says the setup runs on GB300 NVL72 systems with Quantum-X800 InfiniBand networking, and it also points customers to its Secure Agent Workspace reference design for governed deployments with controls around identity, network access, credentials, and runtime policy. The big picture here is that frontier AI competition is increasingly about deployment: where models run, how efficiently they run, and how easy they are to govern inside real organizations. Next, TechCrunch reports that Ashton Kutcher is leaving Sound Ventures, the firm he co-founded with Guy Oseary 11 years ago, to launch a new VC firm with Morgan Beller. The reported focus is early-stage investing in AI infrastructure, energy, and deep tech. That makes this more than a celebrity-investor item. Sound was an early backer of companies including OpenAI and Anthropic, and this move suggests more investor attention may be shifting from the model layer to the compute, power, and engineering systems underneath it. TechCrunch says the split does not appear to reflect trouble at Sound, and that Kutcher will remain an adviser to the firm. For research, a new July 1st arXiv paper looks at a practical bottleneck in medical AI benchmarking: if you use open-ended clinical answers instead of multiple choice, can an LLM reliably grade those responses the way a clinician would? The paper is titled “Clinician-Level Agreement Without Clinical Caution: LLM Evaluator Limits in Medical AI Benchmarking.” The researchers introduce a German open-response benchmark called MedQADE with 3,800 items, annotated by ten practicing physicians and nine LLM evaluators. Their top evaluator reached agreement close to the physician ceiling, with kappa at 0.694 versus 0.709, but the paper says that similarity breaks down on caution. Physicians became more likely to abstain on harder items, while the frontier models assigned definitive scores every time, and the study also found lineage-dependent bias in how models scored related architectures. Bottom line: an LLM judge can look clinically aligned on paper without showing the restraint that makes human medical judgment trustworthy. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ... In headlines, TechCrunch reports that Cloudflare is giving AI companies until September 15 to separate web crawlers used for search from those used for AI training and agents. Companies that do not comply could be blocked by default on many publisher sites. And Ars Technica reports that the U.S. has lifted curbs on Anthropic’s advanced Fable and Mythos models. In the hydrated story text, Anthropic says Fable 5 is now available globally, and that U.S. organizations had access restored to Mythos 5 starting June 26. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. If you enjoyed this episode, don't forget to subscribe, rate, and leave us a review! And that's your briefing for today. Full source links are in the episode notes, and we'll be back tomorrow with what's up next!