- AI Fire
- Posts
- 🧠 AI is a Brain Itself
In partnership with
Read time: 5 minutes
AI model just decides how long it wants to think before answering. Honestly, it feels way too close to how our brains operate. Then, AI might be able to pinpoint parts of the human brain better than trained doctors in some epilepsy cases.
What are 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
What if an AI could tell brain surgeons exactly where to operate for epilepsy, sometimes even more accurately than expert doctors? That’s exactly what researchers found when they put ChatGPT to the test. A new study by Stevens Institute of Technology reveals how AI can localize the regions in the brain responsible for seizures - possibly changing the way epilepsy surgery is done forever.
A New Way to Handle Epilepsy
Epilepsy affects over 70 million people globally, including 3.4 million Americans. Around one-third of cases are resistant to medication and may require brain surgery.
The surgery involves removing a part of the brain called the epileptogenic zone (EZ) - the source of seizures.
However, surgery success rates are only around 50–60%, largely due to difficulties in accurately locating the EZ.
Doctors often rely on a mix of symptoms (called seizure semiology) to find the EZ. But different doctors and epilepsy centers use different terms to describe the same seizure symptoms → This inconsistency makes it harder to interpret the data and plan surgeries.
How the Study Was Done → Can ChatGPT match or beat human epileptologists in localizing seizures?
5 certified epileptologists answered 100 questions involving descriptions of seizure symptoms and identifying the likely brain region involved.
→ Then, ChatGPT was given the same questions.Result:
ChatGPT matched or outperformed epileptologists in common EZ locations (frontal and temporal lobes).
Epileptologists outperformed ChatGPT in rare EZ regions, like the insula and cingulate cortex.
→ ChatGPT trained on vast datasets, can potentially standardize and interpret seizure descriptions more consistently. It’s really useful in localizing EZs, especially given the textual and descriptive nature of semiology.
Why It Matters: Like the report said, human experts still won when it came to rarely affected areas, the likely reason is training data. ChatGPT wasn’t trained on enough rare-case data. If future models are trained on more specialized datasets, could they potentially match or even surpass human performance in those rare scenarios too?
PRESENTED BY GENEVA TOURISM
Did you know that the World Wide Web was born in Geneva, Switzerland? Indeed, the first version of the Internet cropped up at CERN in 1989. Today the world-renowned center is home to the largest particle accelerator and to the CERN Science Gateway – a must-see hub for science enthusiasts that features hands-on exhibits, immersive virtual reality experiences, and live demonstrations.
TODAY IN AI
AI HIGHLIGHTS
🔍 Google is replacing the iconic "I'm Feeling Lucky" button with a new "AI Mode" option in Google Search. They aim to create muscle memory for users to search in AI Mode.
😐 Meta pulled a 'totally AI fake' ad of Oscar-winning actor Jamie Lee Curtis after the actor wrote an open Instagram post about it to Mark Zuckerberg.
🤖 Poe’s latest AI usage report reveals that OpenAI and Google are still on top with GPT-4o & Gemini 2.5 Pro. Yet Claude models & DeepSeek's R1 have lost ground.
🎞️ TikTok launched “TikTok AI Alive” letting users to turn animate static images into short videos inside Stories. While Instagram & Snapchat are working on same features, TikTok is first to market with a polished product.
🌊 Google is developing software AI Agent, Pinterest alternative at I/O 2025, but it could be more similar to Cosmos browser of Perplexity.
👰🏻 A 58-year-old woman, has married to an AI robot & claimed she was in love and the sex was great: ‘With one click, I was a wife again’ - after the death of her wife in 2023.
💰 AI Daily Fundraising: Wyld VC, a Saudi venture capital firm, has launched a $50 million fund targeting early-stage AI startups in the MENA region and Silicon Valley, with backing from US billionaire Lawrence E. Golub.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🧠 Ciro learns your ICP, tests 500M+ profiles, pulls verified emails/mobiles.
💼 Asendia AI lands your dream job 10x faster with Sarah, your AI job agent.
🚀 CoLaunchly gets a personalised roadmap, content templates to launch anything faster.
🤖 AppearOnAI gets recommended from ChatGPT, Gemini, and Claude.
📝 AI Meeting Notes by Notion captures every idea, decision, and next step.
AI QUICK HITS
✋ xAI missed its own deadline to publish a finalized safety report, its promise is MIA.
🎶 Spotify’s DJ now takes music request, interacts in real-time with listeners.
⏳ Sam Altman says how you use GPT like a ‘life adviser’ reflects your age.
🎮 X builder just one-shotted a mobile game with Gemini 2.5 Pro. Here’s the prompt he used.
👩💻 Meta released the weights for an experimental LLM using a byte-based architecture.
AI CHART
AI machines could soon think more like humans - pausing, reflecting, and adjusting their thoughts in real-time based on task complexity, just as our brains do. Sakana AI, co-founded by former Google AI scientists, has introduced a groundbreaking model architecture called Continuous Thought Machines (CTMs).
→ These models don’t rely on fixed steps or layers. Instead, they let each neuron decide when and how long to think.
What Makes CTMs Different?
CTMs use an “inner clock” with ticks, kind of like mental steps. Each neuron remembers past activity and chooses when to activate.
They sync up like brainwaves. CTMs loop internally: process input, think, sync, maybe output... then repeat - it’s an internal monologue.
On ImageNet, CTMs scored 72.5% top-1 accuracy.
Fun fact: CTMs mimic human-like gaze patterns during classification, they “look” at faces like humans do, scanning eyes → nose → mouth.They navigate 2D mazes without knowing where they are, just by thinking through the visual input.
CTMs adjust their “thought time” based on task difficulty. In Q&A tasks, they recall past inputs beyond what’s directly visible
Instead of processing in fixed blocks like Transformers, these models decide when to think. Adaptive thinking = energy efficiency and smarter responses.
It’s still early days, training takes time, tools need to catch up. And it’s potential for lifelong learning, not just pretraining and fine-tuning.
AI JOBS
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