- AI Fire
- Posts
- 💀 GPT-Born AI is a Dead End
In partnership with
Read time: 5 minutes
A Turing Award winner just said what most of Silicon Valley won’t: LLMs like ChatGPT might never reach AGI. And they are a dead end! GPT-6 is going nowhere!
What are on FIRE 🔥
LEARNING PARTNER AIRCAMPUS
You don’t need another productivity hack, You need a system that works without you.
While most people juggle tabs and chase tasks, a few are handing it all off - to AI agents that don’t assist, they act.
Imagine waking up with your inbox cleared, content posted, and calendar perfectly managed, all handled quietly, without you lifting a finger.
On 30th September, Tuesday at 10 AM EST, we're pulling back the curtain.
It’s a Masterclass showing you how to architect AI agents that actually run your work - across 40+ tools - so you can focus on the parts only you can do.
If you're ready to step into the next version of how you work, now’s the moment.
🎟️ Save Your Seat - and let the systems take it from here.
AI INSIGHTS
While everyone else is scaling GPT-6 to the moon, Richard Sutton - a literal AI legend and the godfather of reinforcement learning - just won the Turing Award (basically the Nobel Prize of AI)
In a new interview with Dwarkesh Patel, Sutton said: LLMs like ChatGPT are fundamentally flawed. Here’s what LLMs don’t do:
No real goals + Can’t be surprised
Don’t learn from consequences. Just mimic… without understanding
He proposes a new architecture called OaK. It learns on the fly & doesn’t need massive pre-training → Scaling LLMs won’t get us to AGI. Ever. Which would mean:
The trillion-dollar race to GPT-7, GPT-8? Maybe irrelevant
“Imitation” is a dead-end. Action + feedback is the path forward
The goal-less nature of LLMs, the thing that makes them safe-ish, might also make them dumb-ish (in the AGI sense)
AGI is not bigger GPTs. It’s smarter loops, agents that learn like we do: trial, error, memory, and motivation.
Why it matters: LLMs got us far… but they might not get us all the way. Giving an AI real goals (instead of passive text prediction) is how you get AGI, and also how you might get risk. LLMs are safe because they’re passive.
PRESENTED BY 1440
Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.
TODAY IN AI
AI HIGHLIGHTS
🚀 OpenAI dropped 220 real‑world tasks for public testing, expanding beyond 44 jobs into messy, interactive work. If you’re an expert yourself, here’s the link to participate.
📚 Alibaba-backed Moonshot AI’s Kimi assistant offers ‘agent mode’ for creating multi-page websites, editable slides from simple user prompts. See it in action here.
🎯 OpenAI just rolled out its biggest ChatGPT ad push ever, going beyond the Super Bowl debut with TV, streaming, billboards & influencer spots like these real-life uses.
🚨 Trump posted (& deleted) an AI‑generated Fox News clip touting “medbeds” that can cure any disease, a secret miracle beds he’d “release”. Here’s the video.
💸 If you're new to n8n, we found a guy built 30+ AI agents with n8n & tracked all dollar spent. He shared this post about the brutal cost truth that nobody talks about.
⚡ OpenAI already 9× its compute in 2025 but plans a 125× jump by 2033, more power than India uses today (1.4B people). You wanna see that wild chart? Here it is.
💰 AI Daily Fundraising: Nscale raised $1.1B Series B, Europe’s biggest, from Nokia, Nvidia, Dell and others to build AI “factories” with 100,000 Nvidia GPUs.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🖼️ HunyuanImage 3.0 delivers realistic in-image text creation
🍌 Banana Prompts gets the exact prompts of trending AI artwork
📸 HeadshotCam turns your selfies into professional headshots
📱 Seodisias is a free SEO crawler & web analyzer to grow your site
AI QUICK HITS
🚨 Elon’s xAI sued OpenAI, accusing it of stealing trade secrets
💸 The crazy AI boom spending isn’t like the dot-com bubble
🤝 UAE president meets OpenAI CEO to discuss AI collaboration
🎶 Spotify deleted 75m+ AI tracks using spam filtering systems
🎟 Meta reportedly in talks with Google to explore using Gemini
AI CHART
DeepMind just released the official paper on Veo 3, and it confirms what users were already sensing: video models are reasoning. If GPT-3 turned text into a playground for reasoning, Veo 3 did the same for video.
→ A new theory called “Chain-of-Frames”, the visual cousin of “Chain-of-Thought” in LLMs, revealing how video models can think across time, not just space. This means you can prompt Veo 3 to:
Solve mazes
Simulate physics (yes, like Jenga blocks falling over)
Edit selfies into headshots
Animate scenes from a hand-drawn sketch
Even restore blurry images, all zero-shot
They tested it across 18,000+ videos and built a four-level framework to measure what it can do: Perception → Modeling → Manipulation → Reasoning
Video models can do the jobs of dozens of computer vision tools all in one prompt. LLMs used to be niche. Now they’re everywhere.
It’s not perfect yet. But just like GPT-3 in 2020, the shape of the future is now obvious.
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