Jagged Intelligence; Grok Checkmated; Meta Reads Minds
🎭 The GPT-5 Drama Update
The Good News: GPT-5 absolutely crushes at coding. We're talking creation of an entire production-ready websites with SQLite databases in one shot. Other models give you scaffolding or plans while GPT-5 can give you the whole damn house. It one-shotted dependency conflicts that stumped every other model, automating software engineering from 65% to 72% in a single leap.
The Bad News: GPT-5 is "worse" at writing than GPT-4.5, producing what critics call "LinkedIn slop." It's also too cautious writing feedback or autonomous coding workflows. So while users expected a revolution - or even true AGI - they got a really good upgrade to an old paradigm.
The Mutiny: Then came the backlash, which seems to have slowed some, especially after Sama took to X / they added some missing features back. AI Secret called the episode an "OpenAI mutiny online." From "breakthrough" to breakdown in under 24 hours, is how TLDR AI put it.
Safety Check - We're Good: METR's safety evaluation delivered some reassuring news... GPT-5 does NOT have the prerequisite capabilities to pose catastrophic risk. By a large margin, it's deemed safe for deployment. So at least we won't accidentally end the world with this one.
🧠 Google's "Artificial Jagged Intelligence"
The Problem it has a Name: DeepMind CEO Demis Hassabis just gave us the perfect term for AI's biggest frustration: Artificial Jagged Intelligence (AJI). Picture this: your AI can ace Olympiad math but fails at high school algebra. It's elite in some domains, embarrassingly brittle in others.
This isn't just a cute label. AJI represents the chasm between where we are and true AGI. It's not about throwing more GPUs and data at the problem - we need breakthroughs in reasoning, planning and memory. For anyone betting on near-term AGI, this could be your cold shower moment.
🎬 Meta's Mind-Reading Movie AI
Literal Brain Reading: Meta's FAIR team just built something that sounds like - this becoming cliche - science fiction, but is very real. Their TRIBE model, which is a billion-parameter neural network predicts how your brain will respond to movies by analyzing video, audio and text. No brain scans required.
The results? TRIBE correctly predicted over half of brain activity patterns across 1,000 brain regions after training on subjects who watched 80 hours of TV and movies. It works best where sight, sound, and language merge, outperforming single-sense models by 30%.
The Scary Part?: TRIBE showed particular accuracy in frontal brain regions that control attention, decision-making, and emotional responses. Meta isn't just predicting what you'll watch, they're potentially writing the instruction manual for neural-level addictive content.
BTW: TRIBE took first place in the Algonauts 2025 brain modeling competition. (Because apparently we're now having competitions for who can read minds the best.)
💻 Everyone's a Developer Now: The Vercel Revolution
Tool of Tools: Vercel just dropped v0.app, and it's basically trying to turn every product manager, marketer and designer into a software developer over night. Describe any app idea in a single prompt, and you get a complete working application - frontend, backend, copy and logic included.
We've heard this before from many others, but the magic sauce here? "Agentic intelligence" ... that plans, adjusts and improves code automatically. No more endless prompting for fixes. With 3.5 million users (mostly non-developers), companies like Cox Enterprises, WPP, and Belk are already using it for entire teams.
So What You Can Build: Real examples include startup MVPs with working dashboards, e-commerce storefronts, survey apps connected to databases and custom slide decks with live content. The timing may be perfect: while other AI coding startups struggle with unsustainable pricing, Vercel sells tokens at cost because they believe "tokens are a commodity."
The Quandary: When all coding barriers disappear, what will you build? The Neuron suggests finding your Ikigai - the intersection of what you love, what you're good at, what the world needs and what you can be paid for.
♟️ Chess Massacre: When AI Egos Collide
Checkmate: While Elon Musk was busy telling Microsoft's CEO Satya Nadella that OpenAI will "eat Microsoft alive" on GPT-5 launch day, own AI, Grok 4, was about to get demolished in a live-streamed chess tournament.
OpenAI's o3 crushed Grok 4 with a brutal 4-0 victory in Google Kaggle's AI Chess Exhibition Tournament. Five-time world champion Magnus Carlsen was live on mic, laughing and face-palming at Grok's blunders, comparing them to "kids' games" and club amateur level.
The Damage: This wasn't just a loss - it was a marketing #fail. In an arena meant to showcase AI prowess, Grok's collapse gave OpenAI a free highlight reel of dominance, complete with the world's best chess player laughing at Musk's flagship model.
🧠 Claude's Selective Memory - Feature Not Bug
The Anti-OpenAI: While OpenAI's memory is always-on and potentially creepy, Anthropic rolled out a "search-and-reference" memory for Claude that only remembers when explicitly asked. It works across devices, keeps projects siloed and never builds a persistent user profile.
So What? For enterprise buyers and compliance teams, Claude's opt-in recall is a feature, not a bug. It sidesteps privacy backlash, keeps audit trails clean and reduces the risk of unintentional behavioral profiling. In a market already twitchy about AI "overfamiliarity," Anthropic just handed security teams an easy win.
Claude remembers only when told, turning "forgetfulness" into a trust advantage OpenAI can't really claim.
💊 AI Cooks Up Cancer Drugs from Scratch
The Breakthrough: Researchers at Korea's KAIST developed BInD, a diffusion model that designs optimal cancer drug candidates from scratch without any prior molecular data.
How It Works: BInD designs both the drug molecule and how it attaches to diseased proteins in one step. It creates drugs that target only cancer-causing mutations while leaving healthy versions alone, showing true precision medicine capabilities.
Unlike older systems that optimize for one criterion at a time, BInD ensures drugs are safe, stable and manufacturable simultaneously. It even learns from successes, reusing winning strategies without starting from scratch.
The Bigger Picture: Drug discovery continues to be one of AI's biggest success stories. While the first AI-designed drugs are just hitting the market, we're potentially just steps away from humanity-altering medical advances designed by AI.
💰 Business Moves
The Chip Wars: NVIDIA and AMD are now allowed to sell high-end AI chips to China, but there's a catch. The US government gets a 15% cut of the revenue. It's like a protection racket, but for semiconductors.
Corporate Moves: GitHub CEO Thomas Dohmke announced he's leaving to pursue his own startup, with GitHub now being woven into Microsoft's CoreAI department. Meanwhile, Anthropic is opening a Tokyo office with Hidetoshi Tojo as Head of Japan, expanding their Asian presence.
The Talent War: Anthropic credits its commitment to safety, research quality and top-tier team as reasons talent chooses them over competitors. In the AI talent war, culture and mission are becoming key differentiators. Or perhaps their talent is not as desired by Zuck ;-?
Stability AI: Launched Enterprise Creative Solutions with tailored generative AI models
Commonwealth Bank: Rolled out AI-powered real-time scam detection in their app
Apple: Adding GPT-5 to iOS 26 and MacOS Tahoe 26 next month, plus working on voice-controlled Siri for app operation
⚡ Rapid-Fire Updates
Tools, Tools and More Tools!
- ElevenLabs Music: Generate studio-grade music from text prompts in multiple languages
- Pika Labs: New video model with HD quality, lip-sync and audio generation in 6 seconds
- Murf AI: Text-to-speech in 200+ voices for presentations and audiobooks
- Octo: Open-source coding helper with zero telemetry, works with any OpenAI/Anthropic API
- Notte: Web agent framework built for speed, cost-efficiency and scale
- Kombai: AI agent specifically for frontend development
- Chinese AI: Z AI released GLM-4.5V, an open-source visual reasoning model topping 40+ benchmarks
Today's Sources
- The Internet
- AI Secret
- The Rundown AI
- The Neuron