- AI Fire
- Posts
- ⚰️ Google Veo 3.1 Buried Veo 3.0
Sponsored by
Samsung just dropped a tiny AI model that outsmarts models 10,000× bigger, yes, even Gemini and o3-mini. And Harvard’s AI doctor just made medical history.
What are on FIRE 🔥
PRESENTED BY OTIO
Struggling with research and writing? Meet Otio AI, the smarter way to work.
Trusted by over 200,000 professionals, researchers, and students worldwide, Otio helps you summarize documents, chat with PDFs, videos, and links, and write or edit with AI assistance.
You can even generate PDFs. Whether you're tackling complex workflows or need quick insights, Otio streamlines your tasks with cutting-edge AI models like GPT-5 and more. Save time, stay productive, and focus on what matters most.
Ready to transform the way you work? Try Otio AI today!
AI INSIGHTS
In a direct challenge to the “bigger is better” trend in AI, Samsung’s tiny AI model goes head-to-head with trillion-parameter giants… and win?! → It outperforms OpenAI’s o3-mini, Google’s Gemini 2.5 Pro, and others on the hardest reasoning benchmarks.
TRM is built on a dead-simple idea: instead of one giant pass through data, it loops over its own answer, refining each time until it gets it right.
→ It’s like Chain-of-Thought… without the chain. Here’s what it beats (yes, really):
Sudoku-Extreme: 87.4% (vs. ~55% for o3-mini / Gemini)
Maze-Hard: 85% (beats larger models)
ARC-AGI-2: 8% (still a difficult benchmark for all)
But wait, it’s not a ChatGPT clone. So it won’t summarize your emails. But it will outreason most LLMs on abstract tasks. TRM won’t replace GPT-4. That’s not the point.
You can try it now on GitHub, run Sudoku benchmarks, or build your own recursive agents on top.
Why it matters: It proves that there’s another path forward that doesn’t require a billion-dollar GPU cluster. TRM is basically a counter-punch to the “scale solves everything” mindset.
PRESENTED BY WING
The assistant that scales with you
Great leaders don’t run out of ideas. They run out of hours.
Wing gives you a dedicated assistant who plugs into your workflows – calendar, inbox, research, outreach, and ops – so execution never stalls. Wing can:
Onboard in days, not months
Run the day-to-day so you don’t have to
Adapt as your business grows
With Wing, you buy back time without adding headcount. More focus for strategy, more follow-through on priorities, and a lot fewer “forgot to send” moments.
TODAY IN AI
AI HIGHLIGHTS
🤖 If you’re not using OpenAI’s Agent Builder and think it’s “just another ChatGPT thing,” read these 50 real use cases boosting productivity across almost all fields.
🃏 ChatGPT, Claude, Gemini, Grok & DeepSeek will battle in the first-ever hilarious AI poker showdown on Oct 27. And you can watch every hand in real time on this site.
📑 DeepLearning AI founder just released a new course on Agentic AI & how to build advanced AI agents using tool use & multi-agent collaboration. Get started here.
🔮 We found an old video of Steve Jobs from 1985 showing him seemingly predicting AI tools like ChatGPT. It’s making waves across social media. Watch it to see why.
🧩 Google just launched Gemini CLI Extensions (2 days after OpenAI's GPT "App”). But these can be published by anyone to plug tools: Figma, Stripe & Nanobanana.
🔥 Google is cooking Veo-3.1 (spotted in the “coming soon” section of Higgsfield). Looks like they got early info or access. You can join the waitlist here.
💰 AI Daily Fundraising: Macquarie Asset Management is investing up to $5.0B in Applied Digital to boost AI infrastructure in North America.
AI SOURCES FROM AI FIRE
🎉 Today, let's shout out to our legends of August 2025! You’ve just unlocked a 3-month pass to AI Fire Academy → yep, that means full access to 500+ workflows, tutorials, and AI deep dives:
hlop**@*******valleytech.com / matim*****@gmail.com
thek***@hotmail.com / jar****@gmail.com
NEW EMPOWERED AI TOOLS
💬 AnyStory grows your authentic LinkedIn with only 5 mins/day
⚙️ Prescottdata builds your personalized agent in 10 minutes
🌐 Instantsite gets a live website just by describing your idea
🚀 Synapt spots viral social posts early & replies to grow 3× faster
AI QUICK HITS
AI CHART
Harvard Medical School’s Dr. CaBot became the first AI system to go head-to-head with a human expert in NEJM’s legendary medical case series.
And it did break down its entire thought process like a top physician. Here’s how Dr. CaBot works:
You give it a complex medical case like symptoms, history, labs → Then it builds a slide deck and talks through its reasoning:
Lists the possibilities (aka differential diagnosis)
Rules out distractions and red herrings
Cites real clinical papers
Reaches a final answer
All in about 5 minutes with “uh,” “you know,” and doctor-like phrasing baked in. It’s… kinda freaky right? You can actually watch these AI case talks online. There are already 15 available.
But when I see it’s powered by OpenAI’s o3 model & trained on 100+ years of Mass General’s CPC cases, I’m like “ah, that’s it”.
This is the first time NEJM has published an AI diagnosis alongside a human’s and didn’t shy away from its flaws either. Still, Dr. CaBot isn’t ready for hospitals yet.
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