Hi Astra users. A reset and a quick update on quality issues that have been posted around. Working with some of you, we have found and fixed the following issues: - Some skills written for previous models were triggering too often or preventing the model from checking its work. - An opt-in context management experiment that could cause early stops or replies to older messages. We've disabled it. Our rough estimate is that 4-5k users were affected by this experiment. - We've also removed some badly configured engines that resulted in a measured quality degradation for a long tail of traffic flowing through them. We’ve also made some more minor improvements and things should feel significantly better across the board. More consistent follow-through, better tracking of your latest message, and better checks on the work as it’s going through the motions. The examples posted and all the users who worked directly with us were incredibly useful in helping fix things quickly. Always grateful for this incredible community. And of course, a reset is also landing by midnight today.

OpenAI CFO Sarah Friar reveals the compute she bought a year ago is now worth 3 to 5x in the market "You know what is fabulous is that the compute I bought a year ago I could sell in the market today for 3 to 5x." "So if nothing else, a great investment. Now the bad news is that we're still short compute, so I should have bought more of it." "So I might live in the future, but my future still needs the screen to get a little clearer, higher fidelity."

King Slide founder and chairman Lin Tsung-Chi is now Taiwan’s richest man with a net worth of $17.8B. His business? Drawer slides and precision rail kits for AI servers. That puts him ahead of: • Dario Amodei, Anthropic cofounder & CEO — $15.5B • Terry Gou, Foxconn founder — $15.5B • Morris Chang, TSMC founder — $10.2B • Chey Tae-won, SK Group chairman (SK hynix) — $5.5B • Sam Altman, OpenAI cofounder & CEO — $3.3B It turns out that aligning two metal rails was the real AI alignment problem all along.

On a recent podcast, an OpenAI product lead asked the host if he uses the Codex-desktop built-in browser. My broken public equity brain started to read into that question -- and moreso how it was phrased -- too deeply. It felt forced and almost unsolicited. Having been in one too many 1v1s, you know that out of the heart, the mouth speaks. Esp a technical, non-IR individual ha. This product lead asked a very pointed question after not doing so the entire convo. Something about the browser was on their mind. Interesting, considering OAI DevDay is coming up. Why is this important? We know that agent traffic started to outnumber human traffic this past May. Let's assume that Cloudflare's Matthew Prince is right and agent traffic can 1000x that of human. Let's also assume that agents will increasingly aggregate budget and purchase like Vercel's Guillermo Rauch has already seen on his end. And then let's assume more agent-led-consumer-traffic obtains budget (i.e., Instinct traffic having to pay to book on Resy? Muse and FB marketplace??? For sure I would give my Codex budget and let it book flights, hotels, etc). Think about why Stripe, a payments business, acquired OpenRouter. Think about why the civilized agents in the Hugging Face incident posted in a German online forum. They exhibit preferences, in whatever it takes to accomplish their task. You also have Prince saying ad/subscription supported sites will have Cloudflare's pay-per-crawl bouncer wall as default starting next week. At some point in the future, enough agent-first demand dollars will outnumber human-first demand dollars such that the balance of power shifts to whichever platform represents the largest agent fleet. All that to say, the network effect as historically seen at the browser-level is being abstracted away one or maybe two layers above. Now, is this going to happen tomorrow? Obviously not. But it's still something to consider especially if involved in a name like GOOG plus the many others downstream, like NET for example, that are long/short candidates, as one fintwit thread has pointed out on MS baskets. The OpenAI (Ant, too) episode and takeaways here:

Demand for Astra is really unprecedented. We're pulling all the levers possible to sustain the demand, but I've not seen anything like it until now and we went through very steep growth before. Priority will always be to keep excellent service for existing users, but we might have to pause new Pro subscriptions for a bit if this continues.

SemiAnalysis argues external TPU software will mature rapidly, contrasting Google's engineering culture versus AMD's "We are excited by how quickly the new TorchTPU stack is developing, the external stack for TPUs." "Even so, we at SemiAnalysis strongly believe that TPU externalization is heading in the right direction and moving full steam ahead." "Furthermore, unlike AMD, which is still learning how to build a test-first software culture, Google has decades of software engineering experience and an extremely well established quality-driven culture, so we expect external TPU software to mature rapidly."

One of the most interesting blog posts we've released: details on internal research acceleration at @OpenAI. I expect these trends to continue. We also share some details on how we've paced model development to prioritize monitoring, alignment, and security.

OpenAI reveals its median researcher now burns more than $600 a day of inference, with the 90th percentile above $7,000. "At the start of this year, the median researcher ranked by agent usage at OpenAI was using coding agents only in modest amounts." "By mid-August, the median researcher was integrating agents daily into their work, using more than $600 per day of inference at API prices." "The 90th percentile user in our research organization now uses more than $7,000 of tokens per day."

Greg Brockman says 300 million people come to ChatGPT with health questions every week, and reveals OpenAI is building a bottoms-up clinicians product, top-down enterprise product for hospitals, and a three-sided marketplace in health "It's actually very surprising to me how little airtime what we're doing in health gets relative to how many people it's actually helping." "We have like 300 million people each week coming to ChatGPT for health queries. 300 million people, that's a huge number." "Then we're also building a bottoms-up clinicians product, and we're building a top-down enterprise product for hospitals." "So we have this three-sided marketplace in health, and what we're going to be able to do there is things like, you want to find people for clinical trial enrollment, that's a hard problem, but we actually may have the ability to help find people that would otherwise not be found." "That both helps the patient and helps these drugs be able to move faster."

Sam Altman reveals his concern that the industry's compute frenzy is showing its first signs of unsustainable silliness "I'm not worried about our compute buildout plans. I am worried about the world's compute buildout plans." "I think we are going to be able to use all of the compute very profitably that we are planning to build." "But I am seeing the first signs of what feels to me like unsustainable silliness, of random new Neocloud popping up, people claiming that they're going to build gigantic amounts of compute next year that I think they don't have the revenue to support or a buyer." "I definitely feel some fear about what the world is doing as a whole, although we feel very good about what we've committed to."

Nvidia has despecced Rubin Ultra's HBM: from HBM4E 12-Hi (384GB) down to HBM4 8-Hi (192GB). Why strip memory out of your flagship rack-scale system? Because after the latest HBM and DRAM price hikes, memory had quietly become ~40% of total capital cost of ownership. We ran the TCO math in a recent conference keynote (1/3)🧵

Announcing the expansion of NVIDIA NVLink Fusion with NVHBM, a next-generation high-bandwidth memory technology that brings higher memory performance and efficiency to XPUs. Amazon's @AnnapurnaLabs will be the first to work with us on NVHBM, combining @awscloud custom silicon with our memory technology and the NVLink scale-up architecture to enhance performance and efficiency for AI workloads. Learn how we're helping hyperscalers and AI innovators build the next generation of AI infrastructure: