- AI Fire
- Posts
- 💸 $250/mo to Rent Math God, Google?
Read time: 5 minutes
Google just dropped a math AI trained like an Olympiad gold medalist and it might be too powerful for comfort. It helps with bio, chem, even nuclear-level knowledge.
What are on FIRE 🔥
IN PARTNERSHIP WITH SECTION
Supercompanies are coming. They build AI-natively, will be AI-enabled, and will upend the current business models of incumbent companies.
On August 14, join Section CEO, Greg Shove, as he unpacks how leaders should prepare to pivot and how founders can find themselves among the supercompanyclass.
AI INSIGHTS
Google quietly launched something wild. It’s called Deep Think - a new AI variant inside the Gemini app - based on the same model that earned a Gold Medal at this year’s International Math Olympiad (IMO).
This version isn’t as powerful as the full research version (which can spend hours solving math proofs), but it still performs at a Bronze Medal level, better than most high school math geniuses.
So yeah… Google just gave your phone an Olympiad-tier math brain.
Deep Think uses what Google calls “parallel thinking techniques.” It explores multiple approaches at once before locking in the best solution.
But don’t expect it to replace physicists just yet. Ah, it’s only available to AI Ultra subscribers. Price: $250/month (yes, really)
Google’s own red flag: It might be too smart. Deep Think triggered internal safety alerts:
In model evaluations, Google said it might have reached a “critical capability level” for helping bad actors with bio, chem, or nuclear weapons knowledge.
They're not saying it crossed the line but it’s close enough that they added extra usage monitoring and safety restrictions.
Why It Matters: ByteDance just launched Seed-Prover, a more specialized IMO-grade math model (Silver Medal), and its code is open. OpenAI is expected to drop its own Gold Medal math AI soon. But this round? Google shipped first.
PRESENTED BY GROUPON
Groupon brings you thousands of local steals—from family-friendly museums and outdoor adventures to cozy wine tastings and hottest concerts—discounted up to 70%. Browse once and you’ll be planning three weekends ahead, confident that you’re squeezing every drop out of fun out of the season, and not your wallet
TODAY IN AI
AI HIGHLIGHTS
♟️ Google's DeepMind & Kaggle launched the Kaggle Game Arena, where AI models compete in strategy games. It kicks off August 5 with 8 frontier models: GPT, Claude, Gemini, Grok, DeepSeek,... Watch the livestream here.
🕵️ Perplexity is accused of sneaky AI scraping even when sites say “No” with millions of requests per day. Cloudflare de-listed Perplexity, but it still refuses that action!
🎬 xAI officially dropped Grok Imagine. You’ll turn any simple text or image into a 15-second video, with native audio. But it's quite AI-generated. See some viral examples.
📱 Google's Pixel 10 ad directly mocks Apple’s delayed Siri AI rollout. If Apple is still “coming soon,” maybe it’s time you “just change your phone.” Here’s the funny ad.
🧠 A new trending paper from ByteDance says it solved 5/6 problems on the 2025 International Math Olympiad. So we have Gemini, ChatGPT and now this Seed-Prover!?
🚀 D-Wave Quantum launched a cool AI toolkit, already usable. Even a working image-generation demo here. They will present this toolkit at The AI Research Summit 2025.
💰 AI Daily Fundraising: Observe Inc. raised $156 million to boost its AI observability platform. The company processed over 150 petabytes of data monthly with assistance from OpenAI.
AI SOURCES FROM AI FIRE
🔥 Ep 18 Tooldrop with 5 unfairly good AI tools:
Automate any web task by combining AI prompts
Write 200+ pages of research & books with refs
Get you to #1 on ChatGPT, Claude, Perplexity, Gemini,…
Note: These exclusive resources & reviews are available only in our AI Fire community. You guys can freely ask for support or share personal experience during testing there. Get your full breakdown here (no hidden fee)!
NEW EMPOWERED AI TOOLS
📚 SciSpace AI agent automates research with 150+ academic tools
📝 Verbite delivers publish-ready, high-ranking content with AI Agents
⚙️ Tregno builds & executes a full SEO strategy for you in minutes
🎬 Golpo is the most powerful AI video generator for explainer videos
AI QUICK HITS
🧘♀️ ChatGPT now reminds you to take breaks if you’re in a long chat.
🧠 Google's healthcare AI invented a fake brain part & no one noticed.
🖼️ A 20B-parameter text-to-image model aced official evaluations. Try it here.
🔒 Google’s AI bug hunter just found 20 real security vulnerabilities.
🔋 Google agrees to curb power use for AI data centers during peak times.
AI CHART
Two new charts paint a clear picture: if your job is repetitive, it’s at risk. If it’s human-facing, it’s (mostly) safe.
First, we’ve got the 15 fastest-declining jobs between 2025–2030, based on employer forecasts from the World Economic Forum.
Then, we’ve got the jobs safest from AI, ranked by how much public interaction and human decision-making they require → Together, they show exactly where AI will disrupt the labor market, and where it won’t (at least not yet).
What’s the trend? Anything clerical, repetitive, or tied to legacy systems is being phased out. Automation and AI are replacing these roles with bots, self-checkouts, or predictive systems.
But hold on, not every job is doomed. Jobs with real-time decision making, emotional intelligence, and physical-world problem solving are safe, for now.
But don’t get too comfortable.
Even McDonald’s AI drive-thru failed.
Even Amazon’s “Just Walk Out” tech used humans in India to verify purchases.
→ So a lot of “AI automation” still has humans behind the curtain.
It’s sorting jobs into 2 categories, and you want to be in the “hard to automate” column.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply