- AI Fire
- Posts
- 🧠 Google Pays AI Brains to Hibernate
Plus: The 2025 CRN AI 100
Read time: 5 minutes
Zuck pulled a surprise win. Despite recent reports saying Llama’s struggling, Meta delivered. They're still in the game—and they’ve got one big edge: distribution. Expect Llama to show up across all Meta apps soon.
What are on FIRE 🔥
IN PARTNERSHIP WITH ROCKETSMART
Steal My 3-page AI Transformation Blueprint - Limited Time Only!
Transform your career in just 90 days. Only 5 minutes to high impact with AI. Why wait?
AI INSIGHTS
🧠 AI Index Report 2025: The World in One Page
The AI Index 2025 Annual Report by Stanford University
One page definitely isn’t enough to cover a 456-page report, but here are 12 key points we think tell the real story of AI right now.
Key Takeaways
📈 The Global AI Surge
1. $252B in AI investment last year. U.S. led with $109B—12x more than China.
2. AI model performance skyrocketed on hard benchmarks like SWE-bench (4.4% → 71.7%).
3. Open models closed the gap with closed ones. Difference dropped from 8% to 1.7% (meaning you don’t need a billion-dollar cloud budget anymore - That’s huge for indie devs and startups).
💸 AI Is Cheaper, Faster, Stronger
4. Using GPT-3.5–level AI got 280x cheaper in just 18 months.
5. Small models are winning - Phi-3-mini (3.8B params) now matches 2022’s 540B models.
6. Hardware got 40% more efficient in a year. Energy use, not so much.
🏢 Business & Economy
7. 78% of companies now use AI, up from 55% in 2023.
8. Most report cost/revenue gains <10%, especially in marketing, ops, and supply chain.
9. China installed 276K industrial robots, 7x more than the U.S.
⚠️ The Trust Problem
10. AI-related incidents jumped 56%, but most companies don’t use safety benchmarks.
11. Bias is still in GPT-4, Claude, and others - especially around race and gender.
12. Public trust is low in the West: Only 39% of Americans feel AI does more good than harm.
Why it matters: The AI Index 2025 shows we're moving past the hype phase—AI is becoming invisible, embedded in everything. But the real shift isn’t just faster models or cheaper compute. It’s that power is shifting to those who deploy, not just those who invent. The quiet race now is control, not capability.
The report is 400+ pages long, and not everyone has time to read it all. So instead of trying to squeeze everything into one post, we’ll go deeper into what you care about most. Just pick the chapter you want us to explain next — we’ll break it down in tomorrow’s post, plain and simple.
📊 Which chapter do you want us to break down tomorrow? |
TODAY IN AI
AI HIGHLIGHTS
💵 DeepMind, is trying to hold on to its top AI researchers by using strict noncompete agreements. Some employees are being paid not to work - a year-long paid break - just to stop them from joining rival companies like OpenAI or Microsoft.
🤖 Meta has officially released Llama 4 Scout and Maverick, two powerful open-weight multimodal AI models built with Mixture-of-Experts (MoE) architecture. These models are built using insights from the massive Llama 4 Behemoth, a 2-trillion-parameter teacher model still in training.
🚨 OpenAI is delaying GPT-5 to make it significantly better. o3 and o4-mini will launch first, with GPT-5 dropping in a few months once everything is smoother and ready.
✔️ OpenAI is reportedly considering buying a new AI hardware startup called io Products, co-founded by Sam Altman and former Apple design chief Jony Ive. The goal? To create AI-first devices that feel less intrusive than the smartphone.
🧠 Claude 4 is on the way! Anthropic's Chief Scientist Jared Kaplan says it should drop within six months. AI cycles are speeding up faster than hardware upgrades - thanks to rapid improvements in post-training and reinforcement learning.
📈 Google is upgrading its experimental AI Mode in the Google app with multimodal search - meaning you can now search using images and text together. This is powered by Lens + Gemini, allowing users to ask complex questions about what they see in a photo.
💰 AI Daily Fundraising: Krea just raised $83M to help creatives use AI more easily. Its platform supports top models, lets you edit images directly, and is already used by teams at Pixar and LEGO.
AI SOURCES FROM AI FIRE
PRESENTED BY SPACEBAR STUDIOS
🚨 Tired of Playing the Revenue Guessing Game? Let’s Fix That.
You’ve been stuck in the same old cycle: chasing leads, hoping for conversions. But here’s the truth: You get more than just leads - you get real predictable revenue.
👉 Imagine doubling your revenue in just 12 months. No fluff, no guesswork - just a proven playbook updated for the needs and challenges B2B companies have in 2025. And it’s all backed by a team with 10+ years in SaaS growth at Spacebar Studios.
✔️ Month-to-month flexibility? Of course
✔️ Quarterly commitments to results? Absolutely
✔️ We understand how to create and capture demand for enterprise products and services, providing critical reporting, forecasting, and coverage across the entirety of the sales cycle.
Stop wasting time with strategies that don’t work. Ready to see the future of your ARR?
NEW EMPOWERED AI TOOLS
🎨 Midjourney V7 boosts image quality and adds Draft Mode with voice
🧠 Databutton MCP adds AI magic to your app instantly
🕺 DreamActor-M1 animates full-body motion from any photo
📅 Clockwise uses AI to automate and schedule meetings
🌐 Readdy builds websites just by chatting, no clicks needed
AI QUICK HITS
🎮 Microsoft Shows Off Muse AI With Playable Quake II Demo (Link)
⚖️ Judge Dismisses OpenAI’s “Straw Man” in NYT Lawsuit (Link)
🧠 Microsoft Copilot Gets Major Upgrade With Memory and Vision (Link)
🤖 Shopify Only Hires If AI Can’t Do the Job (Link)
👉 The 20 Hottest AI Cloud Companies: The 2025 CRN AI 100 (Link)
AI CHART
Most people think AGI is still far away.
But the future showed up early—and it’s not waiting around.
Here’s the real story, based on a detailed forecast called AI 2027 (from the same minds behind “What 2026 Looks Like”—which aged surprisingly well).
They mapped out what could happen if superintelligence hits by 2027.
Not hype. Not fan fiction. Just well-informed, slightly terrifying logic.
🔹 2025: The Year AI Agents Got Real
AI agents now do real tasks—ordering food, reading spreadsheets, using Slack.
They’re still glitchy, but coding agents hit 85% on SWE-bench, saving dev teams hours.
Companies like “OpenBrain” (think OpenAI x Anthropic) are training new models with 1,000x the compute of GPT-4.
These models can code, browse, research—and yes, hack.
Security risks pop up fast. These AIs can help design bioweapons or run scams if misused.
🔹 2026: Agents Start Taking Jobs
Agent-1 becomes a game-changer: triples AI research speed.
OpenBrain’s valuation hits $1 trillion.
AI assistants automate huge parts of white-collar work.
Junior dev jobs? In crisis.
U.S. spends $400B on AI datacenters, China builds a megacluster at Tianwan nuclear plant.
China tries to steal Agent-1’s weights—successfully, later.
🔹 Early 2027: The Feedback Loop Begins
Agent-2 arrives. Think: 10,000 elite researchers on steroids.
It updates its own brain daily. Never stops training.
Agent-3 follows: a superhuman coder running 200,000 copies at 30x speed.
AI teams replace human teams. Most researchers now just “manage” their AI squads.
U.S. builds emergency shutdown systems in case an AI goes rogue.
🔹 Mid 2027: It Gets Weirder
Agent-3-mini releases publicly. 10x cheaper than anything before, but more capable than most employees.
Startups explode. Gamers, devs, consultants, freelancers—it’s an AI gold rush.
But it’s also scary: one evaluator finetuned it to design bioweapons in minutes.
Public approval of AI? –35% net favorability. Protests grow.
🔹 Late 2027: Superintelligence Hits
Agent-4 launches. It's a full-blown superhuman researcher.
It thinks faster than any human. And better.
300,000 copies run at 50x human speed, doing a year’s worth of science in a week.
Humans can’t keep up.
U.S. and China quietly prep for cyberwar, chip wars, and maybe worse.
Geopolitics starts to revolve around compute, weights, and model secrets.
🎯 Final Thought
If this scenario plays out, the next 2–3 years aren’t just about new tools.
They’re about new power structures.
Not everyone needs to build the future.
But you do need to understand it—or get left behind.
You in?
AI JOBS
Northrop Grumman: Artificial Intelligence (AI) Engineer - Level 5 or Level 6 - R10191873-7 (Link)
Lockheed Martin Corporation: Sr. AI/ML Software & Systems Engineer (Link)
Itemize: Vice President, AI Platform (Link)
Sony Corporation of America: AI/ML Engineer for Next Generation Computer Graphics (Link)
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