- AI Fire
- Posts
- 👑 22 of Top 50 AIs are Chinese
In partnership with
Read time: 5 minutes
ChatGPT’s once-clear lead? Fading fast. A hacker used Claude to run a full-blown cybercrime spree. And a new benchmark just crushed most “smart” research agents.
What are on FIRE 🔥
LEARNING PARTNER AIRCAMPUS
💭 What if you had 4 AI Agents working for you while you sleep?
An agent that answers calls + emails 📞
An agent that manages your daily tasks 🗂️
An agent that runs your social media 24/7 📲
An agent that automates workflows end-to-end ⚡
In just 3 hours, during our live masterclass on Friday, August 29th at 10:00 AM EST, you’ll learn to build your own AI Agents Army - with zero coding.
✅ Save 20+ hours a week
✅ Scale your business without hiring
✅ Automate your personal + professional life
AI INSIGHTS
For a long time, ChatGPT felt untouchable. But according to a16z’s latest AI trends report, that lead is shrinking fast. Gemini, Grok, and even some low-key Chinese players are gaining momentum:
Google’s Gemini is now the #2 AI app on mobile and a 90% share on Android.
Grok’s a top-5 AI tool on the web and ranked #23 on mobile, with 20M+ users and growing.
Meanwhile, Meta is stuck at #46 on the web, not even ranked on mobile.
DeepSeek dropped 22% on mobile, 40% on web
Claude is holding on with steady web growth, but flat on mobile
Perplexity? Quietly climbing in both places
China’s AI makers are making global plays and they’re everywhere:
Doubao (ByteDance) is now #4 mobile AI app
Quark (Alibaba) ranks #9 on web
22 of the top 50 mobile AI apps are made in China
Lovable and Replit, two vibe-coding startups, made the list for the first time. Apps like PixAI, Talkie, Seekee, AI Mirror, and Photo AI are right on the edge of mainstream.
Why It Matters: This is the first report where ChatGPT looked… surrounded. It still leads, but not by a mile anymore. Gemini is climbing. Grok is exploding. Claude is steady. And mobile is where the real AI war is happening now.
PRESENTED BY ROKU
Find your customers on Roku this Black Friday
As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. To that end, Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting options. After all, you know your customers, and we know our streaming audience.
Worried it’s too late to spin up new Black Friday creative? With Roku Ads Manager, you can easily import and augment existing creative assets from your social channels. We also have AI-assisted upscaling, so every ad is primed for CTV.
Once you’ve done this, then you can easily set up A/B tests to flight different creative variants and Black Friday offers. If you’re a Shopify brand, you can even run shoppable ads directly on-screen so viewers can purchase with just a click of their Roku remote.
Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.
TODAY IN AI
AI HIGHLIGHTS
🧠 To see which AI is actually reliable, The Washington Post ran a 900-answer test for 9 tools from ChatGPT, Claude, Google, Perplexity, Meta & Grok. Here's the test.
💥 Ad maker PJ Ace did it again. Months after going viral with an AI ad for Kalshi (100M views), he just made another for David Beckham’s company IM8. Here’s the super-realistic video.
🤝 Anthropic and OpenAI just audited each other’s AI models for safety risks → a first-of-its-kind collab. You can read Anthropic’s findings on OpenAI (and vice versa) here.
📊 Anthropic analyzed 74,000 educator chats with Claude & released a new report at how AI is used in higher education. Turns out, professors are now quietly adapting too.
🌨 Google DeepMind’s experimental weather AI just nailed its first real-world test, predicting Hurricane Erin’s path more accurately than traditional forecasting methods.
💰 AI Daily Fundraising: FieldAI raised $405M to build one universal robot brain that works across all robot types. Their physics-based AI cuts risk, and is already used in energy, delivery, and construction.
AI TUTORIAL
Tavus lets you create lifelike AI humans that see, hear, speak, and respond instantly. And you don’t need a PhD or a dev team to launch one. You can spin up branded, emotionally-aware agents in hours, not months. Behind every Tavus agent:
✅ Face-rendered visuals and voice
✅ Real-time emotional intelligence
✅ Memories, roles, and personality
✅ Full brand control with white-labeling
✅ Easy plug-in to any stack
✅ Instant global reach, 24/7
Whether you need 1 or 10,000, Tavus scales with you. Now it’s time to turn AI into a human experience.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🍌 Gemini 2.5 Flash Image with "nano-banana", SOTA image model
⚡ CodeX builds webapps in minutes for free with 27+ AI models
💬 Onepard turns your content into a friendly chat page. No code
📸 ScreenshotReports creates branded reports from screenshots
AI QUICK HITS
📢 Here’s a viral system prompt for ChatGPT to speak like a real person
🎥 Alibaba just updates its video AI model with “film-quality avatars”
📈 Google Workspace’s offering a new Free AI video tool to Google Vids
💼 Stanford found AI killed 13% of jobs for 22–25 y/o in coding, support
👨💻 A hacker used Claude to run a full-scale cybercrime hitting 17 orgs
AI CHART
AI2 just dropped AstaBench, a rigorous new benchmark suite built from the ground up to actually test scientific AI agents with 2,400+ real research tasks.
AstaBench is designed around 5 brutal truths about agent testing:
Tasks must be real-world and hard
Tool use must be controlled
Costs must be tracked, accuracy isn’t everything if you’re burning $$$
Standardization is key → same tools, same formats, less dev bias
You can’t claim progress without knowing what you beat
They tested 57 agents across 22 architectures. Here’s the high-level result:
Asta v0 (their own agent): 53.0% → uses 5 LLMs and 5 task-specific sub-agents
ReAct + GPT-5: 43.3%
Data analysis is still weak, no one scored above 34%
Literature understanding is strongest, 44 agents passed at least one benchmark
GPT-5 helped general agents (like ReAct), but oddly hurt specialized ones. So it might’ve been tuned specifically for ReAct-style workflows.
This is the benchmark to watch if you're building AI tools for research, medicine, engineering or anything where trust, reproducibility, and cost matter.
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