- AI Fire
- Posts
- 🧠 Gemini Gets a New Job
In partnership with
Google’s AlphaEvolve just gave Gemini a new job: finding better algorithms for science, chips, logistics, and power grids. The bigger question is what happens when AI starts improving AI itself.
What's on FIRE 🔥
IN PARTNERSHIP WITH HUBSPOT
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
AI INSIGHTS
Google shared new results from AlphaEvolve, its Gemini-powered coding agent for designing better algorithms.
The simple version: it doesn’t just write code. It improves the systems behind science, infrastructure, and business.
Key results:
Genomics: Cut DNA variant detection errors by 30%.
Power grids: Improved feasible solution rates from 14% to over 88%.
Quantum: Found circuits with 10x lower error for Google’s Willow quantum processor.
Google infra: Helped improve next-gen TPUs, cache systems, Spanner, and compiler storage.
The business side is growing too. Klarna doubled training speed for a major transformer model, FM Logistic improved routing by 10.4%, and Schrödinger saw around 4x speedups in molecular modeling.
The bigger signal: AI agents are moving beyond chat and coding help. AlphaEvolve shows how AI can improve the hidden algorithms running chips, grids, logistics, science, and AI itself.
PRESENTED BY NEO
Stop switching apps. Your browser can do it all.
Every tab you open, every copy-paste into ChatGPT, every lost train of thought — that's your browser failing you. Norton Neo fixes it. Built-in AI works directly inside your session. Hover to preview. Search everything from one bar. VPN and ad blocking included, free.
AI SOURCES FROM AI FIRE
1. 12 Claude Code Tips & Tricks to Level Up From Beginner Setup to Advanced Agent Teams. Discover 12 Claude Code tricks to build apps, AI agents, websites, and workflows faster, including hidden workflows most builders still miss.
2. Video: 6 NotebookLM Pro Tips to Save You 20+ Hours of Research (99% users missed THIS). These are tips I use to automate my deep research, build a "Digital Brain," and 3x my productivity.
3. Video: The Only Codex Guide You Need in 2026: Skills, Plugins, Automations... My goal: take messy notes, email workflows, project files, then use Codex to turn them into useful outputs to actually open & reuse.
4. Video: I Connected Claude to Meta Ads: Build Ads, Check Pixel, and Plan a $5/Day Test Faster. The goal: use Claude to help draft campaigns, write ad ideas, check tracking, and review the strategy before anything goes live.
FIRE RECAP: BIGGEST AI NEWS THIS WEEK
🛰️ Anthropic signed a massive compute deal with SpaceX, giving Claude access to Colossus 1 with 220,000+ Nvidia GPUs and 300MW+ capacity. The AI war is now about chips, power, and who gets compute first.
⚡ OpenAI made GPT-5.5 Instant the new default ChatGPT model. It claims 52.5% fewer hallucinated claims than GPT-5.3 Instant on medicine, law, and finance prompts.
☁️ Anthropic reportedly committed $200B to Google Cloud and AI chips over five years. That makes Google even more central to Claude’s future as the frontier model race gets more expensive.
🏛️ The Pentagon signed classified AI deals with major players like OpenAI, Google, Nvidia, Microsoft, AWS, and SpaceX, while Anthropic was left out. Big AI is now moving deeper into national security.
🧭 Google shut down Project Mariner, its web-browsing AI agent, on May 4, 2026. The tech is moving into products like Gemini Agent and AI Mode, likely before bigger agent updates at Google I/O.
TODAY IN AI
AI HIGHLIGHTS
🦙 Ollama just patched a critical bug called Bleeding Llama. If exposed online, attackers could leak API keys, system prompts, code, and chat data from local AI servers.
🔗 Google is adding more website links inside AI Overviews and AI Mode. New sections like Further Exploration and Expert Advice should send more users back to original sources.
💾 The Wayback Machine is getting squeezed by AI again. Large hard drives used for web archives are now up to 3x more expensive, while the archive grows by 100TB per day.
⚡ Cloudflare says AI made 1,100 roles obsolete, even as revenue hit a record $639.8M quarter. CEO Matthew Prince said AI usage jumped 600% in 3 months.
🕵️ Stripe CEO Patrick Collison says token theft is becoming a serious AI problem. Fraudsters now make up 1 in 6 new signups, burning free AI credits at machine speed.
💰 AI Deals: Anthropic signed a $1.8B, 7-year cloud computing deal with Akamai to keep up with surging demand for Claude. The deal is the largest contract in Akamai’s history, as Anthropic races to secure more compute after reporting 80x growth in annualized revenue and usage in Q1.
NEW EMPOWERED AI TOOLS
🌐 Codex in Chrome lets the Codex app control your browser to navigate websites, fill forms, and complete tasks using your active logins.
📊 How AI-pilled are you? gives your company an AI Fluency score in ~12 minutes, with benchmarks and recommendations inspired by teams like Shopify, Zapier, and Ramp.
🧑💼 Prism helps companies hire beyond traditional recruiting networks by finding high-fit candidates across the open talent market with faster, personalized outreach.
⚡ Zappy by ZapDigits turns marketing data from Google Analytics, Search Console, and more into client-ready AI reports and answers in seconds.
AI BREAKTHROUGH
Bigger context windows help AI read more. But Ctx2Skill shows the harder problem: AI still struggles to learn from new documents.
On CL-Bench, even top models are weak:
GPT-5.1: 21.1%
Claude Opus 4.5: 21.0%
Gemini 3 Pro: 15.8%
So the best models still fail most tasks when they must learn new rules, workflows, or hidden patterns from unseen context.
Ctx2Skill fixes this by turning messy documents into reusable natural-language skills. No fine-tuning. No human labels. Just a short Markdown-style skill file added to the prompt.
The result: GPT-5.1 jumps from 21.1% to 25.8%. GPT-4.1 jumps from 11.1% to 16.5%, even beating Gemini 3 Pro’s 15.8% baseline.
The key lesson: Future AI agents won’t win by stuffing more PDFs into prompts. They need systems that turn company docs, workflows, and research papers into clean skills they can reuse.
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