AGI is here, here's what we know

Welcome back! Nvidia's CEO declared AGI on Sunday, and OpenAI's chief scientist picked the same day to ask the whole industry to slow down. The chips behind Astra are why your Xbox costs $150 more. Anthropic answered the biggest week in AI with a 400-year-old math problem, and GTA VI fans got tired of waiting for Rockstar.

In today's Generative AI Newsletter:

Nvidia: Who called AGI, and who at OpenAI asked for the brakes the same day?

Consumer Reports: Why does your Xbox cost $150 more?

Anthropic: What did Claude finish in 11 days that took humans 358 years?

GTA VI: Who's building the game before Rockstar ships it?

AGI is here, here's what we know

Nvidia's CEO said it on Sunday . AGI has arrived. Jensen Huang built the 100,000 chips Astra trained on, and he's adding 400,000 more.

You've scrolled past a hundred AGI posts. This one's different because OpenAI backed it with its own numbers the same day.

OpenAI's AI now does the research that builds the next AI. Its researchers get 3.1 days of machine work per human day, and the median one burns $600 a day in tokens.

Altman's target was an AI intern doing days of research alone by this September, then a full AI researcher by March 2028.

The intern arrived on schedule. That leaves 18 months to AI that improves itself.

Claude got left in the dust for the first time in a while. Astra beats Fable 5.1 on 17 of 19 tests OpenAI scored both on.

The AI intern is worth talking about a little more. It’s not a human intern using AI. It’s a full-on AI intern, no humans involved. This means AI will keep upgrading and improving itself very soon, which is the literal definition of AGI.

Doesn’t need instructions. Can teach itself. Keeps improving no matter what.

Then the chief scientist spoke.

Jakub Pachocki wrote that no one is prepared for what's coming, no lab has solved alignment well enough to keep going at full speed and he wants the whole industry to slow down. Altman reposted it and called it important.

The man running research at the fastest lab on Earth asked for the brakes.

The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.

This guide distills 10 AI strategies from industry leaders that are transforming marketing.

Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI

Learn how AI-driven emails achieved 94% higher conversion rates

Discover 7 ways to enhance your marketing strategy with AI.

Outpace your competitors by mastering AI.

Get Your Free Guide

AI ruined tech for everyone

Every gadget you wanted this year costs more. Consumer Reports added it up .

Xbox Series X up $150 to $800.

PS5 up $100 to $150.

MacBooks and iPads up $200.

The cheapest smart home Echo Dot went from $50 to $80.

New car carries thousands of chips now, so add $2,000 to $3,000.

The memory inside all of it is going to AI data centers instead. Samsung, SK Hynix and Micron now send about 70% of their capacity to the chips Astra trains on.

Prices for the ordinary kind jumped as much as 98% in three months.

Memory costs what it did 20 years ago. AMD brought back a 3-year-old chip because it runs on cheaper memory, and Apple put an old iPhone chip in its $699 MacBook. Tim Cook told the Wall Street Journal the situation is unsustainable.

Let’s hope it stops soon before your average devices become too expensive to own.

Anthropic's response to AGI is average at best

Anthropic's answer to the AGI weekend was a math problem from 1637. Fermat scribbled it in a margin, it took 358 years to prove and Andrew Wiles' proof ran 129 pages that took months to check.

The math world has spent two years trying to make a computer verify every step of it. Claude did it in 11 days , mostly on its own.

13 million lines of proof, 29,500 smaller theorems along the way, five times the size of the entire library mathematicians built by hand. It's still a research post.

Read that again. It didn’t solve anything or discover something new. It proved it.

Seems like a response to ChatGPT Astra. This isn’t the first time Claude does something great with maths. Previous achievements were even more mesmerizing.

People are expecting a better model from Anthropic. Not new math achievements. The sad news is the new model they have planned is just another version of Opus.

Polymarket has it at 82% for this month with no announcement behind it.

A proof and a betting market, against the week OpenAI just had. Average at best.

GTA VI fans stopped waiting for Rockstar

GTA VI has been delayed twice, and it's been 13 years since GTA V. So fans, according to Polymarket, are building the game themselves from the released screenshots with GPT-6.

@petergyang made four games in 2 hours with Astra, a Star Fox shooter, a first-person shooter on a moving train, a StarCraft level and a deck builder, on 20% of his $200 plan.

@MatthewBerman got a playable Fall Guys clone from one prompt.

OpenAI's own launch demos include a kart racer and a house modeled in Blender and turned into an Unreal Engine walkthrough.

Building a game meant learning an engine. Now it means describing one, and ChatGPT hosts it for you.

GTA VI is due November 19. That's the official date, assuming it holds.

Tool of the Day: Rosebud AI

Rosebud AI turns one sentence into a playable game inside a browser tab, with the engine, art tools, sound and hosting all in the same place. You pick which model builds it, Claude Opus is marked as the best option, and the output is real JavaScript you can open and edit. Publishing gives you a link anyone can play on a phone, multiplayer comes wired in and you keep 100% of the profits on paid plans. 2.5 million games made so far, free to start.

Try this yourself:

Go to rosebud.ai , describe your game in one sentence or pick a template.

Switch the model picker to claude-opus before you hit Create.

Playtest in the same tab and describe changes in plain English.

Hit Publish and send the link to a friend.

Who it's for: anyone who read the GTA story and wants to try it without touching Blender.

Everything else you shouldn't miss

Trump was asked if AI could go rogue and launch cyberattacks : no, he said, we watch it very carefully, we're leading China in AI, and whoever wins AI wins it all.

The US and China will hold their first AI safety talks in mid-September : Reuters says Treasury Secretary Bessent leads, joint monitoring of AI cyberattacks is on the agenda and the White House denies any meeting is planned.

Google mapped every connection in a male fruit fly's brain : 166,000 neurons and nearly 11,700 cell types stitched from millions of 2D images, the first complete male map after the female one.

Anthropic's IPO prospectus could land this week : the FT reportedly says the filing comes as early as this week, with Morgan Stanley and Goldman leading and a listing in September or October.

Learn more about AI from the experts building it

📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.

📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.

🐦 Follow us on X for breaking AI news and real-time industry updates.

🧠 Learn how to build your next AI application with practical resources and expert guidance.

🎓 Start learning with free AI courses from GenAI Academy.

💰 Invest in the GTM infrastructure of the AI economy.