📉 AGI Mark is Being Halved!

Stanford’s Viral GPT Tips to Copy. . 

Free AFIRE Guide | AI Academy | Advertise | AI Mastery A-Z

ai-fire-banner

Plus: Build A 24/7 Custom AI Assistant That Works While You Sleep

Read time: 5 minutes

One of the biggest AI skeptics just admitted AGI could be 5 years away (previous 10 years). And Korean researchers built an AI that creates cancer drugs without any data.

Start Listening Here: Spotify | Apple Podcasts, YouTube

LEARNING PARTNER AIRCAMPUS

Build the Future Before Anyone Else

Imagine having AI experts running your calendar, inbox, socials, and documents — all at once, all on autopilot.

On Tuesday, 15th August at 10AM join a select group of innovators who are redefining how work gets done with AI.

In this 3-hour AI Agent Mastery Session, we’ll show you step-by-step how to design, deploy, and automate AI systems that run entire workflows without you lifting a finger.

Whether it’s managing projects, handling customers, generating content, or analyzing data, these agents will work across every platform you use.

No code. No tech jargon. Just a proven blueprint for turning your ideas into powerful AI-driven results — faster than you ever thought possible.

AI INSIGHTS

🚀 One of AI’s Biggest Skeptics Just Halved His AGI Timeline (in half)

François Chollet – the creator of Keras and long-time critic of AGI hype – just revised his forecast for artificial general intelligence from “10 years away” to “5 years away.”

And no, it’s not because models are getting bigger. It’s because they’re finally getting smarter… in the way that actually matters.

In a new interview, Chollet says his optimism comes from a fundamental shift in AI behavior. For years, he argued that deep learning models were glorified parrots.

Now? We’re seeing signs of “fluid intelligence”: the ability to adapt to novel problems on the fly, with no prior examples.

That’s a big deal. And it’s exactly what his new benchmark, ARC-AGI-3, is designed to test:

  • The AI is dropped into a simple video game.

  • No instructions. No labels. No examples.

  • It has to explore, figure out the rules, and learn to win, just like a human would

Each AI gets tested on ~100 of these unique game worlds, and success depends on one thing: skill-acquisition efficiency. Humans can pick these up in under a minute. AI models? Not even close (for now).

→ “As long as we can come up with problems humans can solve but AI can’t,” Chollet says, “we don’t have AGI.”

Instead of each AI model acting like a “perpetual intern on day one” (Dwarkesh’s words), any skill learned by one AI becomes instantly available to all other AIs.

Why It Matters: This kind of collective intelligence means AI compounds knowledge exponentially. Humans learn in silos. AIs in this system would learn as a hive mind. Dwarkesh said it best: “That would basically be the singularity.”

TODAY IN AI

AI HIGHLIGHTS

🤖 Stanford’s Jeremy Utley shared a clever ChatGPT prompt that has changed the way they use AI for good. It’s gone viral for a reason, you can also copy that prompt here.

🛠 Latest updates: new GPT-5 mode options (Auto/Fast/Thinking), GPT-4o restored, GPT-4.5 restricted to Pro tier, 3K GPT-5 Thinking weekly rate limit. Full details here.

💵 Claude Sonnet 4’s context window is expanded to 1M tokens but with higher fees. You need to be on tier 4, which means you have purchased at least $400 to use this.

🤖 Adding more fuel to the fire, the Grok reply bot on X chipped in and implied that Musk was a hypocrite. The ChatGPT social account decided to take things up a notch.

⚠ Recently, a man landed in the hospital after following ChatGPT’s advice in his diet. A reminder: you still need to verify AI outputs, especially when it affects your body.

💰 AI Daily Fundraising: AI chip startup Rivos, valued over $2B, seeks $400-500M to develop power-efficient chips & compete with Nvidia. Key backers include Intel Capital, MediaTek,…

AI SOURCES FROM AI FIRE

AI EVENT

Rubrik AI Identity Week – The AI Identity Crisis

Gain critical insights into the latest trends and techniques used by adversaries to exploit identity systems. During this virtual event, we’ll dive into:

  • AI’s Role in Identity Attacks

  • A Live Hacking Demo

  • Proactive Identity Resilience

AI CHEAT SHEET

NEW EMPOWERED AI TOOLS

  1. 💡 TantyAI prompts GPT, Gemini, Grok, Claude… simultaneously

  2. 🎥 MagicLight turns text into up to long-form 30-minute videos

  3. 🧵 Threadly turns any idea into engaging X threads in 3 minutes

  4. 💰 Bio Calls could monetize all your social media in 60 seconds

AI QUICK HITS

  1. 💬 Gemini added Temporary Chats and new personalization features.

  2. 🚪 One co-founder of Elon Musk’s xAI officially departed the company.

  3. 🧠 OpenAI clarified that GPT-5 thinking’s context window is 196k, not 32k.

  4. 🎮 Skywork dropped Matrix-Game 2.0 (like Genie 3) for interactive video.

  5. 🌱 Google invested $9B to expand AI infrastructure in Oklahoma.

AI CHART

💊 AI Designs Cancer Drugs From Scratch. No Data Needed

Korean scientists just dropped a serious breakthrough: a new AI model that designs cancer drugs from zero. It’s called BInD, and it’s not just speeding up drug discovery, it’s rewriting how it’s done.

BInD is a diffusion model that skips the traditional trial-and-error process and generates new drug molecules from scratch.

→ No templates, no prior examples. Just raw, generative chemistry.

Most AI drug models today still follow the usual steps:

  1. Start with known molecules

  2. Tweak them

  3. Test for binding

  4. Repeat until something works

But BInD? It skips all of that to design the molecule and its target-binding interaction in one go → kind of like generating a key and lock at the same time. Even better: it optimizes for multiple properties simultaneously.

And here’s the really wild part: when it finds a working design, BInD recycles success, reusing smart strategies across future generations of molecules to get even better.

And because it doesn’t need training data, it opens doors to treating rare mutations or diseases we’ve never seen before where traditional datasets simply don’t exist.

We read your emails, comments, and poll replies daily

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

 


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *