Hey yall, welcome back to ThursdAI, this is Alex, let me catch you up!
Today on the show, we covered 1 week with Astra (hint, it’s not quite AGI yet despite what we were told), DeepSeek V4.1 catches up to the frontier at a fraction of the cost, and Meta launches a free AI agent with it’s own computer, that will take over the OpenClaw/Hermeses of the world for most people. Also huge this week, OpenAI claimed that a swarm of 10K agents of their unreleased model solved the Navier-Stokes, one of the millennium problems!
I was stoked to have Chris Alexiuk from Nvidia on the show to cover the innovations DeepSeek put into this latest model!
Oh, and the guy who quit Anthropic this week, and wrote an essay about “AI is going to kill all of us” somehow got 130M views on X, a mirriad of TV interviews and rekindled the doomerism movement, we talk about that too!
Also, I already told about FullyConnected, CoreWeave’s premier conference that’s coming up, but they told me about a new announcement today, and you’re not going to believe who it’s about (not AI related). As a reminder, ThursdAI subscribers get a free ticket!
Ok, let’s dive in!
OpenAI claims a Navier-Stokes solution from a 10,000 agent swarm from an unreleased model - with some drama (X, Blog, Paper)
If you’ve been reading ThursdAI for a while, you may remember that a model couldn’t tell which is higher, 9.9 or 9.11 and naysayers said that “AI can’t math”
Well, this week OpenAI claimed that a swarm of 10K agents, of an unreleased model, found a solution to the Navier-Stokes problem, in about 88 hours! They published a huge 160+ page paper and a Lean proof of the solution! I said on the show, this is like the moon landing equivalent of AI doing things humans didn’t do before!
Now, keep in mind, this is only a claim from OpenAI, the Clay Mathematics Institute is still reviewing it (moved the status of this problem from “unsolved” to “under review” so this isn’t independently verified yet) but it’s still an insane deal.
The agents sent 2.7 million messages and burned about 130B tokens, of a model that has no public price yet, so it’s hard to estimate the cost of this run, all for a 1M prize that OpenAI said they will not claim.
The coolest thing I think we got from this paper, in addition to solving on of the most important and hardest problems in mathematics, is this chart above, where OpenAI shows their unreleased model and how much better it is on Math problems compared to... GPT-6! The “AGI” model we got just a week ago. So so much to look forward to.
The drama behind this
I don’t want to get into the drama behind this too much, but if you’ve seen this online, there release wasn’t without it’s hiccups. Apparently, OpenAI caught wind that an a duo of researchers Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic, in personal capacity), have independently solved the related Euler problem and were about to go public. Apparently, the duo used a mix of GPT and Claude to work on this problem. OpenAI caught wind of this and started working on their own solution on September 1st. A day before OpenAI dropped their release, Buckmaster posted that OpenAI is about to drop it, and in communications with him, they offered him to co-author the paper, only if the Anthropic guy is removed. To which he said no.
He then had some claims that maybe OpenAI trained on some of the papers and chats they put into Codex, which OpenAI refuted, while noting “we cannot rule out the possibility that de-identified usage data helped improve the model”.
OpenAI also say that the OptOut toggle works, and neither mathematician provided a screenshot that they opted out of the training, so it’s hard to say what really went in there
My take, I don’t really care. Two years ago, we told you that reasoning is coming and AI is going to be doing superhuman things, and we finally see the first signs of it, this is a problem that no humans was able to solve for over 60 years! Navie-Stokes probably doesn’t change your friday, but there are so many other that they can solve with this approach! Cancer research, room temperature superconductors (remember LK-99? that’s also a search problem) and much more. Kudos to OpenAI for this, and I’m looking forward to see the new heights of mathematics. And for the mathematicians who “disagree” with OpenAI solving or not solving this problem, why don’t you post your own Lean proofs instead of fuming publicly online?
GPT-6 - Not quite AGI, yet?
After a week with Astra GPT-6, and Jensen Huang announcing AGI is here, I think we can do a quick recap
I and all the co-hosts have been Astra-maxxing for a whole week now and the results are in. This model is incredible at coding, it goes very very deep, however, definitely not AGI quite yet. It has a very jagged frontier, it does some things incredibly well (most demos online are building whole games and apps in 3D and those are mind-blowing) but I started seeing many folks go back to GPT 5.6 Sol etc.
I tink some of it also has to do with price, Astra runs faster and is significantly more expensive so token limits for folks are draining fast, but also with controllability, folks are likely still using their old and unoptimized prompts. Speaking of prompts, here’s a great writeup from OpenAI how to rework your prompts (which you can send to Astra and have it review your prompts!) for better results
The AI doomerism has quite a week (Coxon post, Thread, Hubinger, Marks, Christiano, An Alien Mind)
I started ThursdAI with the notion to counter anti-ai and doomerism, and bring positivity to the AI world, so we had to cover this. An Anthropic employee who previously worked at OpenAI, posted on X about leaving Anthropic, saying that both labs are racing towards uncontrollable self-improving superintelligence and that it could be a disaster of the “end all of humanity” kind.
His post sits at 130M impressions (after being basically a nobody on X before) and he’s been interviewed by Fox, AP, Time magazine, WSJ, NBC and a host of senators, Bernie (chief doomer) included, reposted his post on the same day!
The funniest thing is that this post got about 26x more attention than Ilya Sutskever’s post about leaving OpenAI. Just nuts
Within four days, seven current and former employees from Anthropic, OpenAI and DeepMind said in public that they believe that AI could kill us all. Also notable that Paul Christiano, who is one of the most interesting AI doomers out there, has joined the OpenAI foundation, and Daniel Kokotajlo, the famed OpenAI whistle-blower, joined Joe Rogans podcast to talk about AI doom.
Each one of these incidents in vacuum is normal, but having all these happen in a a span of a few days just feels, inorganic. Some folks are even saying that this is a well coordinated doomerism campaign!
I want to be fair to Coxon, folks who worked with him at OpenAI say he’s the real deal, and cares deeply about AI safety and humanity, however, he only worked at Anthropic for 6 weeks before publicly leaving, and now every interview he does sayd “Ex Anthropic employee”.
Whether it’s a coordinated effort or not, it’s still a very important discussion, after the pacingthefrontier letter and the HuggingFace hack incident, and it seems to have made waves. Sam Altman just told staff that he’s not opposed to pausing and have petitioned the US government to regulate AI as well
Look, I don’t disagree that we’re dealing with a very powerful technology, however I don’t believe that scaring the bajeesus out of everyone is the right way to handle this. Politicians use fear to get votes and get elected, they don’t really care about tech progress, and framing this in a way that “we pause or we’re dead” ignores all the good that AI is about to do. Cure cancer, find solutions to climate change, helping solving povery. All these seem like out there ideas but they are coming. The US GDP is already growing at an unprecedented rate and a lot of it is due to AI. In any rate, as I said on the show, I’m not against pausing, just after we solve cancer. Then we can pause and reassess, till then, nobody is telling me how China’s government is going to pause if US pauses, and if they don’t, they will reach superintelligence before we, and I don’t want to live in that world!
Open Source AI
DeepSeek V4.1 Flash: the whale is back, and it’s cheap (X, HF, TokenJuice)
Speaking of.. chinese AI! Deepsek (The whale) resurfaced this week with V4.1 Flash, and don’t let the name fool you, this is not just a .1 small update. 552B with only 8B active on prefill and 16B on decode, 1M context, trained from scratch on 45T multimodal tokens! plus as always, MIT license.
Chris from Nvidia joined us to break it down, and his main point stuck with me: every DeepSeek release comes with one of the best engineering reports you can read, and this one is the most data-pilled they’ve ever done. The paper basically says it out loud, everything else is nice, but it’s the data. 45T tokens isn’t a huge number anymore, but the cleaning they describe goes way beyond what anyone else publishes (Chris said even his own beloved Nemotron’s open pipelines are less thorough).
Yam opened a new corner of the show, “I Told You So”, because DeepSeek went back to an encoder-decoder architecture. Not the old one from before GPT-2, this one has a pile of battle tested tricks that make it work at half a trillion parameters. His verdict after testing it all day: the best open weights model you can host for coding right now, and it’s not even close to the largest one.
This chart is the one to look at. KV cache per token went from 389,000 bytes in the first DeepSeek (Nov 2023) to about 890 bytes now. Nisten did the math live, over 400x smaller. That’s why this model is so cheap to serve, and as Yam kept yelling, we shouldn’t take it for granted, this is the actual moat and they just put it in the open.
Evals, briefly: 90.6 on Terminal-Bench 2.1 (above Opus 5 and GPT 5.6 Sol), 74.2 on DeepSWE 1.1 (also above both), and on an Open Design leaderboard it lands second behind Astra at two cents a task. Not twenty cents. Two. DeepSeek’s own evals, so the usual asterisk applies.
Two more things. Friend of the pod Aaron Batilo (he works on CoreWeave Inference, this is a side project, not sponsored) put up TokenJuice.ai, free and fast DeepSeek V4.1 Flash in exchange for your requests as training data, hosted in the US. Nobody wants free DeepSeek? Go try it. And Nisten had Astra build a 3D visualization of the whole architecture, every weight a cube sized by its bytes on disk, link in the TL;DR.
This Week’s Buzz 🐝: Pitbull is coming to Fully Connected (Fully Connected, CoreWeave Hacks)
Tbh, I was super surprised by this! This isn’t about AI, but exciting non the les Pitbull, Mr. Worldwide himself, is headlining Fully Connected! Yes, really. So the free ticket we give ThursdAI folks is now also a Pitbull concert ticket!
The rest of Fully Connected is still a great reason to come: September 29 to October 1 at Moscone South in SF, 2,000-plus engineers, Sarah Guo from Convitction hosting, Dr. Fei-Fei Li from WorldLabs keynoting, and ThursdAI live from the floor! The code for a free ticket is THURSDAIFC2026, register at the link above. 19 days out.
Before that, CoreWeave Hacks is this weekend, September 12 - 13 in our SF office. The theme is Agent Loops, and the prizes are crazy. Definitely sign up and come hack with us!
Meta launches Muse, a free 24/7 agent with its own computer - the OpenClaw for your mom (and you!) (X, My thread, Site, Security, My YT breakdown)
Meta, after selling back Manus to China, is finally back with a agent product, and this one is actually quite amazing! It’s called Muse, and it’s powered by Muse Spark. It’s really really fast, and oh... it’s free (up to 100M tokens per week?). With it’s own cloud VM, and a browser, it seems like the folks at MSL are going towards taking over the agentic world.
If you’ve installed OpenClaw and moved to Hermes, if you’ve played with Grok Bot (still excellent) and got an Instinct invite, you know the drill. An agent that with your permission can read your emails, browse (and do shopping for you). But Muse is different in a few key ways, not least of which their insane distribution (Facebook, Instagram, WhatsApp, Messenger all having over 2B users)
My first experience with muse blew me away, they have a very strong Stripe Link integration (native) and my Muse was able to get me tickets to a dinner tomorrow (Happy Rosh haShana btw!), finding the hidden link, checking my calendar and booking it, all before Instinct, the other assistant i gave this task to, even replied! It’s really fast!
UX as the differentiator, price as the convincer
Muse is different from the other agents, first of all, because of the price sticker. It’s free! With a very generous 100M tokens, and a beefy cloud VM, this on its own is huge. It’s also very very fast, and to be honest, I didn’t expect this level of polish at the jump!
Meta knows how to launch products, this is VERY polished. Unlike OpenClaw or Hermes which require command line knowledge, Muse is there for you when you sign in with your Meta account, and it’s ready to go to work! Of course, like with everything, the more you share (connectors are there for Gmail, Drive, WhatsApp and tons of other stuff) the more effective and personalized it gets
I particularly like the avatar there, you can choose to customize your own, and it shows what it does (unlike other agents that either just show progress or show you command line outputs), they even generated a few different videos of my wolf doing different things while it does them! Just wonderful.
Self onboarding product and proactive UX
I constantly talk about 2026 is the year of the proactive AI agent, and this seems like one of the first ones where I enjoy their proactive. Not only scanning my inbox and telling me “you got this email, what do you want me to do with it?, Muse saw that my daughter’s birthday is coming up and suggested to plan a party, which I did (it got balloons and a helium tank in Target for me, it’s waiting for me to pick up while I edit this)
These proactive recommendations also make Muse “Self onboarding” in a smart way, especially the “ideas” tab they have on the left that shows folks who never used an AI agent before what they can do wit it. Really really well done
The Zuck shaped elephant in the room : Privacy and Security
But it’s META! Folks on X scream and say they will never share their data with that company, not after everything that happened in the past.
Meta is obviously aware of how folks feel and how much sensitive data they allow these agents to accumulate, and on the security page they outline not only the safety measures they took with Muse (noted is the Sentinel additional process that looks at all your conversations and incoming data to your agents and judged independently if it’s safe of if anyone is trying to hack you), Meta is also offering a bounty of up to $300K to anyone that finds a security issue with Muse.
On the privacy side, the highlight for me is the Confidential VM announcement, while they are still working on this, Zuck shared that Meta recruited Moxie Marlinspike (the founder of Signal) to work on this, he’s the guy who also helped Whatsapp E2E encryption. This should make the virtual machine (that Muse uses to hold all your data) incredibly secure, so much that Meta (verifiably) will not be able to access your content, even if Zuck himself wants to look at your calendar! I’m super stoked by this and hope that this approach will be open sourced and adopted by all the other companies!
Look, I didn’t mean to turn this into a full review of Muse, there’s a LOT more there, like connectors, some of which are proprietary to Meta, and some are novel (like the iphone native ones) and the upcoming 1Password integration.
I posted a full review it on YT and definitely check out that full review, but I think that when Meta launches something that big, you all ought to know what the excitement is about.
LMK if you’ve tried this in comments or if you don’t trust Meta with your data, let me know as well
Instinct - the other agent in the room this week (X)
We didn’t give Instinct too much attention in the show, but it’s also been blowing up lately, an imessage native agent that can do a lot. This week, it got an update with it’s own email you can fwd things to and a cool concept of a “trusted person network” where my Instinct can talk directly to my wife’s Instinct (and to the Instincts of other folks I trust) to sort out plans between us, so the agents negotiate dinner and I just show up. Instinct is doing some very very cool things, and is also free, definitely worth hunting down for an invite (LMK in comments if you wnat one, I have a few left).
Now, the funniest thing about X AI folks, that they wouldn’t trust on of the biggest company in the world with their data, but they will yolo it into a startup by a 22year old 😂
Apple iPhone 18 Pro, A20 Pro, and the AI bits (X, Newsroom)
We mostly skipped the Apple event on purpose. iPhone 18 Pro with the A20 Pro chip (a dual 16-core Neural Engine, double the AI compute of the A19 Pro), Siri AI landing in English beta with iOS 27 on September 14, and the iPhone Duo foldable at $1,999, which has nothing to do with AI even if it’ll be a nice vibe coding machine. The AI parts that actually interested us: AirPods live translation (shout out Milosh in the chat), and the Apple Watch continuously transcribing on device with a “what did they say” button, no cloud. I’ve been using Siri AI. It’s fine. It’s nowhere near the agents above.
AI Art & Diffusion
GPT launches updated Images 2.5: Flare and Sunburst (X, Blog)
Two new API models, Flare for speed and volume and Sunburst for careful editing (with native transparent backgrounds), both at $30 per million image output tokens, and OpenAI claims up to 50% lower latency than Images 2.0. In ChatGPT you get Sketch (type @Sketch and draw), comment-on-image editing and templates.
My hands-on take: I spent the evening making this week’s thumbnails with Images 2.5 through Fal and with the help of Cursor, and the first round was rough. The artifacts were very bad and it make me look really fat.
Turns out that was on me, not the model. My prompts were written for GPT-image-2, full of “8K, cinematic, hyper-realistic” prompt addition that GPT image 2.5 Sunburst treats as “overcook everything”, and the only reference photo we fed it was itself AI-generated.
So I asked Fable to do some prompting research, fixed the inputs, set the quality on high rather than max, and one tiny line that says “real photograph, no heavy retouching”. Second round it beat Nano Banana Pro on both my likeness and the text on the tiles! Very impressive and very fast!
Wrapping up
There is so much more that was released this week, for example, I used the new Cursor “projects” feature with Fable 5.1 as my chief of staff today and it was marvelous, no more just “chats” and it even helped me wrangle my Grok Bot and Muse agents. OpenAI launched their GPT-live-1 model that powers their voice chat in API, so now you can make the same amazing experiences in your own apps as ChatGPT voice mode has, which is absolutely best in class, and both Google and Suno released updated music models (Lyria 3.5 and Suno 6). I’ve added all those things in the notes.
I think this week we saw both ends of the AI spectrum, huge advances in personal (muse) and frontier (Astra) AI, and the rise of Doomerist on the other. I can’t wait to check in to see how next week is doing!
See you then, and if you haven’t yet, a subscription to ThursdAI is free and really helps us to keep going. Thank you!
TL;DR and show notes
Hosts and Guests
Alex Volkov - AI Evangelist, Weights & Biases & CoreWeave (@altryne)
Co-hosts: @WolframRvnwlf, @nisten, @yampeleg, @ldjconfirmed
Guest: Chris Alexiuk, NVIDIA (@llm_wizard)
Open Source LLMs
DeepSeek V4.1 Flash: 552B MoE with 8B/16B active, encoder-decoder, 45T multimodal tokens, 400x smaller KV cache than DeepSeek V1, MIT; free for a limited time on TokenJuice (X, HF, TokenJuice, Nistens visualization)
Desert Ant Labs debuts 18 on-device models with native SDKs, including Voz speech-to-text (X, Blog, HF, GitHub)
InclusionAI Ling-3.0-flash-VL, 124B vision language MoE with 5.5B active, MIT (X, HF)
Big CO LLMs + APIs
OpenAI claims a Navier-Stokes Millennium Prize solution from a 10,000 agent swarm on an unreleased model, with a Lean proof; Clay review pending (X, Blog, Paper)
Jacob Coxon resigns from Anthropic, seven insiders warn of extinction risk in four days, 130M+ impressions (X, Thread, Hubinger, Marks, Christiano, An Alien Mind)
OpenAI says it reached the automated research intern milestone, automated researcher targeted for March 2028 (X, Blog)
Apple debuts iPhone 18 Pro with A20 Pro, Siri AI beta, iPhone Duo, AirPods live translation (X, Newsroom)
OpenAI brings GPT 5.6 Sol and GPT-6 Astra to ChatGPT Voice (X, Release notes)
This Week’s Buzz
Tools & Agentic Engineering
Meta launches Muse, a free 24/7 personal agent on its own Linux VM with native WhatsApp, iPhone connectors, Stripe Link and a Confidential VM roadmap (X, Alex’s thread, Site, Security)
Cursor launches Projects view- one view to rule them all (X)
Cognition ships SWE-2, near frontier coding scores at up to 70% lower cost, free for a month on Devin paid tiers (X)
Instinct, the iMessage agent, adds email and a Trusted Person network (X)
AI Art & Diffusion
Voice & Audio
