A Guy Used AI To Cure His Dog's Cancer*

March 16 , 2026 · Episode Links & Takeaways

Author

AI Daily Brief
March 16, 2026

HEADLINES

NVIDIA GTC Preview: The Groq Chip and the Move to Full-Stack AI

It's a big week for NVIDIA as their GTC developer conference kicks off in San Jose. The big pre-event speculation centers on a new chip system developed in collaboration with Groq (the chipmaker, not Grok the chatbot). NVIDIA acquired Groq in December and is expected to announce the first collaborative product this week: Groq's language processing chips integrated into NVIDIA's rack-scale servers, marking NVIDIA's first serious push at inference. This will also be the first time NVIDIA has manufactured an AI chip outside of TSMC, using Samsung's foundry instead. Patrick Moorhead of Moore Insights summed it up: "NVIDIA is no longer a chip company. The company plans to present itself as a full-stack, heterogeneous AI infrastructure platform spanning training, pre-fill, decode, inference, and agentic orchestration."

AI Agents Are Now a Material Risk in SEC Filings

While software CEOs have been publicly downplaying AI disruption risks, their legal departments are telling a different story. So far this year, 27 firms have listed AI agents as a material risk in SEC filings — up from just 7 this time last year. Figma, Workday, and HubSpot are all on the list, despite their CEOs recently dismissing concerns. Figma CEO Dylan Field said at their last earnings call that unsupervised agent delegation makes you "a very brave person" — and yet Figma's 10-K filed the same day acknowledged agents may "change how people access and interact with digital products in ways that reduce reliance on traditional software applications." The individual disclosures don't say much on their own, but the volume is another signal we've moved past the tipping point on agents.

Seedance 2.0: Hollywood Forces ByteDance's Hand

ByteDance has paused the global launch of Seedance 2.0 after Hollywood studios including Disney, Warner Brothers, Paramount, and Netflix sent cease-and-desist notices. The model launched in China to massive reaction — including that viral Tom Cruise/Brad Pitt fistfight clip — and the Motion Picture Association called it "unauthorized use of U.S. copyrighted works on a massive scale." The holdup now isn't just implementing guardrails but refining them so they don't block too much unrelated content, the same problem OpenAI had with Sora 2. Chinese users are already reporting the model is severely overconstrained, and the foreign release has been pushed back indefinitely while engineers refine the moderation system.

Former Anthropic Researchers Found Neolab for AI Science

A new startup called Mirendil, led by former Anthropic researchers Behnam Neyshabur and Harsh Mehta, is raising $175M at a $1B valuation with Andreessen Horowitz and Kleiner Perkins co-leading. Both founders spent their time at Anthropic working on long-horizon scientific reasoning and automated AI research. The company aims to conduct AI-enhanced scientific research in biology and materials science — a category that's rapidly gathering investment, and one to watch as these new labs scale up through the year.

Google Maps Gets Conversational

Google Maps is rolling out Ask Maps, a Gemini-powered conversational interface for navigating and planning trips. It integrates with Gemini's memory, so restaurant recommendations can draw on what it already knows about your preferences. Google is also launching Immersive Navigation — a new 3D view built from Street View and aerial imagery that depicts buildings, overpasses, and terrain. Once again, Google flexing multimodality and its ecosystem integration as a competitive moat.

ServiceNow CEO: College Grad Unemployment Could Exceed 30%

ServiceNow CEO Bill McDermott told CNBC that AI agents could push unemployment for college graduates "easily into the mid-30s in the next couple of years." The current reality is already concerning: unemployment for recent graduates stands at 5.6%, but 42.5% are underemployed — the highest since 2020. Computer science majors have the highest unemployment at 7%, though their underemployment rate is relatively low at 19.1%. This kind of prediction is precisely what's driving the frenetic AI discourse right now.

MAIN STORY

A Guy Used AI to Cure His Dog's Cancer — And What It Says About AI's Second Moment

This episode is nominally about a viral story involving a dog, a chatbot, and a cancer vaccine — but it's really about the state of the AI discourse right now, and why everything feels so heightened. The discourse is currently at an 11, all the time. On one side: terrifying statistics, viral doom charts, and a cascade of layoff announcements. On the other: normal people managing teams of AI agents and doing things that were never previously possible. The divergence between mainstream perception and actual capability has never been higher — and both are in an incredibly heightened state. The frame that makes sense of this is that we are in AI's second moment.

SHAGGY DOG STORY

AI's Second Moment
"The cloud of dust is proportionally bigger."
The first moment was the ChatGPT moment at the end of 2022. This is the Claude Code / Opus / Codex / agentic systems moment. And just like then, we're getting a proportionally bigger, more dramatic cloud of discourse around it. There are five key differences between now and then worth holding in mind: capabilities are dramatically higher; billions of people are now in the conversation, not just early adopters; economic stakes are real and immediate, not theoretical; AI has become a useful corporate fall guy for unwinding COVID-era overhiring; and all of this is happening against a backdrop of significantly elevated political volatility. There's a sixth difference too: three and a half years of the AI industry doing a genuinely poor job of explaining itself to anyone outside its own bubble.

The Karpathy Jobs Chart
"Wildly misinterpreted, which I should have anticipated."
On Saturday, Kaito on X posted about a weekend project he'd found in Karpathy's GitHub: a treemap scoring 342 BLS occupations on AI exposure from 0-10 using an LLM. Average score: 5.3. Software devs: 8-9. Medical transcriptionists: 10. Twitter instantly flooded with takes treating this as a diagnosis of imminent job destruction. What most people posting about it didn't do was read the actual page, which included a prominent caveat from Karpathy himself: "These are rough LLM estimates, not rigorous predictions. A high score does not predict the job will disappear." Karpathy was so frustrated by the response that he took the whole thing down, noting the exposure was scored based on how digital a job is — which has no bearing on what actually happens to those occupations.

The Nuanced Takes on Exposure and Displacement
"Exposure does not mean threat of displacement."
The Update Newsletter's Stefan Schubert pointed out that the pace of automation doesn't follow from the exposure scores. Chicago Booth economist Alex Imas put it more bluntly in all-caps: exposure can literally mean the opposite of displacement — AI-exposed jobs may see increased hiring and higher wages, depending on demand elasticity and the composition of AI-exposed tasks within a role. Anthropic's head of economics Peter McCrory agreed, noting that even in highly exposed jobs, the remaining bottleneck tasks may ultimately increase demand for complementary human skills. Toronto economist Kevin Bryan went further, saying he'd bet $1,000 that most "susceptible" jobs see an increased share of labor between now and 2030.

The Dog Cancer Story
"One man and a chatbot outperformed the pharmaceutical pipeline."
The other conversation that went viral this weekend was the story of Paul Conyngham, an Australian tech entrepreneur whose dog Rosie was diagnosed with cancer that didn't respond to chemo or surgery. ChatGPT suggested getting Rosie's DNA sequenced and using DeepMind's AlphaFold to find mutations that could be immunotherapy targets. He took that to Pall Thordarson at the University of New South Wales, who developed a bespoke mRNA vaccine in under two months — the first personalized cancer vaccine ever designed for a dog. Vittorio's viral thread framed it as proof that one man with a chatbot had outperformed the pharmaceutical discovery pipeline. The reality, as Thordarson himself clarified on X, is more nuanced: some tumors shrunk but it's not a cure, costs are hard to assess since many people donated time, and veterinary research regulation is quite different from human health. And arguably this is more a story about Nobel Prize-winning AlphaFold than about ChatGPT. Still — AI enabling a layperson to navigate DNA sequencing, drug target identification, and the relevant scientific literature with no prior biology background is genuinely remarkable, even with all the caveats.