AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
AI Solves Open Math Problem & Multi-Silicon Inference Cloud Raises $80M
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's Monday, March 24th, 2026. Frontier AI model has solved an open mathematics problem for the first time. GPT-54PR successfully tackled a problem on Ramsey Hypergraphs from Epoch AI's Frontier Math Benchmark, a collection of problems specifically designed to test whether AI can contribute to genuine mathematical research. The solution was found by researchers Kevin Barreto and Liam Price prompting GPT-5-4 Pro, and it was confirmed by the problem's author, Will Bryan, who called it an elegant solution that eliminates an inefficiency in existing constructions. Here's what makes this significant.6, Gemini 3.1 Pro, and GPT-5 IV's extended thinking mode, can all solve it too. We're seeing frontier models reliably contribute to open mathematical problems now. That's a new threshold. In other news, a startup called Gimlet Labs just raised an$80 million Series A to tackle the AI inference bottleneck in a clever way. The company, founded by Stanford professor and exited founder Zayn Asgar, has built what it calls a multi-silicon inference cloud. The idea is that different parts of an AI agent's workload have different hardware needs. Inference is compute bound, decoding is memory bound, and tool calls are network bound. No single chip handles all of this optimally. Gimlet software slices up agent workloads and routes each piece to the right hardware, whether that's a GPU, CPU, or high memory system. They claim 3 to 10 times speedups at the same cost and power. They've already partnered with Nvidia, AMD, Intel, ARM, Cerebras, and D-Matrix. With data center spending potentially hitting$7 trillion by 2030 and current hardware utilization estimated at just 15 to 30%, there's a massive efficiency gap here. This is a company to watch. Apple announced the dates for WWDC 2026, June 8th through 12th, and for the first time in a while, they're explicitly teasing AI advancements. Last year's conference focused on the liquid glass interface redesign with AI barely mentioned. This year should be different. Apple is expected to unveil a revamped Siri with better personal context and on-screen awareness, and they recently signed a deal with Google to use Gemini for AI features on their platforms. We might also see updates to Apple's foundation model framework for offline AI and expanded integration of tools like ClaudAgent and Codex in Xcode. For a company that's been perceived as playing catch-up in AI, this could be their moment to show they're serious. Alright, a few more things worth knowing about today. LittleBird raised$11 million for an AI recall tool that reads your computer screen and stores everything as text, no screenshots, just searchable context. The founders previously built and sold Sentio to AlphaSense. They argue that storing text rather than images makes the system lighter and less invasive than competitors like Microsoft Recall or the now acquired Rewind. It includes meeting transcription and automated daily summaries. The key question, as one of their investors noted, is whether they can find that one-killer use case. Mozilla AI launched an open source project called CQ, short for colloquy. It's designed as a knowledge sharing commons for AI coding agents. The premise is interesting. Stack Overflow has collapsed from 200,000 monthly questions to under 4,000 as developers shifted to AI assistants. But now those AI agents keep hitting the same problems independently, burning tokens and compute. CQ lets agents share what they've learned across code bases. Mozilla's building it openly and looking for feedback. And a quick one to wrap up a video demonstrating an iPhone 17 Pro running a 400 billion parameter language model locally trended on Hacker News. Running frontier scale models on mobile hardware without cloud connectivity is becoming real. That's it for today. Thanks for listening.