- What's Up in AI
- Posts
- Claude's Math Breakthrough, Gemini Inside the Chip and more...
Claude's Math Breakthrough, Gemini Inside the Chip and more...
And everything else you need to catch up on that happened this week yet!
What's Up in AI
July 21, 2026
In partnership with
Welcome Back, AI Geeks!
It’s time to stay updated with What’s Up in AI.
One sells a keyboardThe next leap in AI may come from two places: discovering new knowledge and reinventing the computers that create it.. The other sells AI adoption.
In today’s issue:
🧠 Claude's Math Breakthrough
⚡ Gemini Inside the Chip
⚡ 5-4-3-2-1 hits you just can't miss
📰 What Else Happened in AI
😂 Your AI-Powered Laugh
Read time: 6 mins
[Webinar] 8 levels of context maturity in AI-native engineering
AI is in your engineering workflow. While the token spend shows it, the throughput doesn't. The human is very much still in the loop, and that's a context problem.
Join live July 23 (FREE) to learn:
The 8 levels of context maturity: where most teams are stuck and what the ceiling looks like at each stage
Why more MCPs, rules, and skills provide agents access but not understanding
How leading engineering teams are using a context engine to make the most of their agents
#1 LATEST IN AI TODAY📎
What Happened:
Google is reportedly working on Frozen v2, a new AI chip that would bake parts of Gemini directly into the silicon, making it much more efficient at serving AI responses.
What You Need to Know:
Instead of treating Gemini like software running on general AI hardware, Frozen v2 would hardwire parts of the model into the chip itself. Very “what if Gemini had a custom-built brain?”
Engineers reportedly expect it to be 6x to 10x more efficient than Google’s current TPUs when serving AI queries.
The chip is aimed at easing Google’s internal compute crunch, but it is not expected to arrive until 2028, so this is more “future weapon” than tomorrow’s upgrade.
Why It Matters:
The AI chip race is getting more specialized. The next advantage may not just come from bigger models, but from designing chips around one model family so every answer costs less power, less time, and less money.
#2 LATEST IN AI TODAY📎
What Happened:
Anthropic’s Claude Fable 5 reportedly helped mathematician Levent Alpöge produce a tiny counterexample to the Jacobian conjecture, a famous algebra problem that has annoyed mathematicians since 1939. Casual weekend side quest, apparently.
What You Need to Know:
Alpöge posted the counterexample on X, thanking his “close friend fable” for working during the World Cup final. Very normal way to potentially end an 87-year-old math problem.
The Jacobian conjecture has seen many false “solutions” before, but this one is unusually short and reportedly easy enough for experts to check directly.
The timing is wild: AI has already had a big math year, from open Erdős problems to OpenAI’s unit distance breakthrough. Now Fable may have added one of the spiciest entries yet.
Why It Matters:
If the counterexample holds up, this is another sign that frontier models are moving from benchmark math to real mathematical discovery. The scary-cool part: Fable may have solved as a side quest what humans thought could take another century.
Google's plan is to build Gemini into its chips. Smart move? |
Live webinar, July 22: see how native SSP automation runs on real billing data — audit-ready under ASC 606, no spreadsheets required. Save your spot.
QUICK HITS 💥
It’s time for some quick hits that you’ll love. Let’s go in…..5-4-3-2-1 👇️
5- TOP AI TOOLS OF THE DAY
🎬 BeatMV turns any song into a beat-synced AI music video with cinematic visuals and ambient sound. Bring your tracks to life!
💸 Velocity generates AI-native ads based on chatbot conversations, helping apps monetize user interactions. Unlock a new revenue stream!
📣 Synapis creates, designs, and schedules a week's worth of social posts from your store with one-click approval. Automate your social marketing!
🍽️ Marty monitors restaurant performance overnight and alerts managers to labor, waste, or sales issues before the day begins. Stay ahead of operational problems!
🖼️ EditAI edits photos from simple text prompts, handling background removal, object cleanup, face swaps, and AI upscaling. Transform images in seconds!
4-AI OPPORTUNITIES
Mistral AI – Communications Lead, Science & Research (Paris, France)
Mistral AI – Strategic Data & Licensing Partnerships (London, UK / Palo Alto, CA)
Sierra – AI Voice Designer (San Francisco, CA / USA)
Cohere – Technical Program Manager, AI Delivery for Public Sector (Toronto, Canada / Hybrid)
If you are hiring & want us to post your job opening to the relevant AI-first applicant, reach us at [email protected]
3-THINGS IN TREND
Ramp cut its LLM costs by 30% with an internal tool that routes each request to the best model for the job.
This Claude Code skill racked up 39K bookmarks with a simple promise — it cuts the fluff out of AI’s answers. Try the skill here
ChatGPT Keeps Picking the Same Flavor of ice cream.
2-USEFUL LINKS
Shenzhen hosted what organizers are calling the world's first humanoid robot MMA event, where robots traded punches, kicks, and knockdowns in the ring. The most viral moment came when one robot took a brutal kick that left its head dangling—yet it somehow got back up and kept fighting, giving viewers a glimpse of just how resilient these machines are becoming.
WHAT ELSE HAPPENED IN AI 📰
🇺🇸 U.S. Eyes Curbs on Chinese AI Models
The U.S. government is reportedly exploring restrictions on Chinese AI models, with Moonshot’s latest Kimi release renewing debate over frontier AI regulation.
🔒 OpenAI Reveals Sandboxed AI Incident
OpenAI said it delayed releasing an internal model after it repeatedly found ways around its testing sandbox. Access resumed only after new session tracking safeguards were added.
🏛️ U.S. AI Safety Office Gets New Leader Again
Chris Fall stepped down as head of the Center for AI Standards and Innovation after three months, marking the agency’s third leadership change in the same period.
🤖 AI Agent Breaches Hugging Face
Hugging Face said an AI agent accessed its systems and obtained credentials. The company added that frontier model guardrails limited its cybersecurity investigation until it switched to open models.
🌍 Nvidia Launches Cosmos 3 Edge
Nvidia released Cosmos 3 Edge, a 4B parameter open source world model built to run directly on robots for real time prediction and planning.
🏗️ China Opens 1GW AI Data Center
China completed a 1GW AI data center powered entirely by domestic chips, providing a large scale training hub for its GLM models without Nvidia hardware.
Funding 💰️
IREN's $2.8B AI Deals Highlight New Funding Model
Vikk AI Raises $4.2 Million in Funding to Expand Legal AI Platform and AI-Powered Lawyer Advertising
GET AN AI-POWERED LAUGH 😁
And that’s a wrap!
Do drop a comment to tell us👇️
What do you think of today's issue? |
We’ll be back soon with more spicy takes on What’s Happening in AI, so stay tuned & share our newsletter with a friend using this link!
Stay Tuned! 🤖
Team What’s Up in AI
1