OpenAI announced a solution to the Navier-Stokes problem, which I had never heard of because I'm not a mathematician, but which is apparently a very big deal:
The question of whether smooth three-dimensional fluid motion can break down has remained unresolved for roughly 90 years.
There is some controversy over whether two human researchers (one at Anthropic) working on similar problems might've accidentally influenced the OpenAI model. It doesn't sound like that's true, but it does feel a little unfair regardless, where credit likely means so much to people.
Did a little Redis housekeeping, cleaning out orphaned records that were sitting around wasting memory. Down to 23 GB of RAM on a 48 GB machine. Still probably too much.
I had a nap yesterday and a late Dr. Pepper, so my sleep was all off. Caffeine really affects me late in the day. Woke up at 2am worried about all sorts of business and social things I can't control.
Let it go. 🎶
Might be about time to get a Switch 2... Ocarina of Time remake looks good. 🕹️
Ben Thompson in today's Stratechery blog post:
AGI is AI that learns continuously. That’s not the case with current large language models.
I agree with Ben's definition of AGI, but LLM memory is so good now, it's almost indistinguishable from learning. My ~/.codex/memories folder is 7 MB, which doesn't sound like much, but it's only small Markdown files. It's incredible how much it knows. That doesn't even count ChatGPT's dreaming memory system and computer history.
How am I just now hearing this interview with Jensen Huang? "You're not talking to someone who woke up a loser." 🤣
This could be really good. Trailer for a new documentary from Nathan Fielder and Lance Oppenheim about Elizabeth Holmes: You Can See Everything. Amazing cut to the wider shot at the end.
Today I updated the Micro.blog web interface with better links and icons in the timeline, and a new context menu for editing your own posts or muting a conversation. There are a lot of related design fixes too. Just a better design and more room for us to build new things on.
Reflecting on the Astra launch... It was a killer business instinct for Greg Brockman to say, "yeah, this is basically AGI." Dario Amodei is much more cautious. He's not going to say that about Fable. But it works because it actually feels true. Marketing not based on truth is just bullshit spin.
I wanted to quote a couple things from this OpenAI blog post on alignment by Jakub Pachocki:
The fundamental challenge of AI alignment is generalization. As machines become smarter, they find themselves working on higher-level concepts, and placed in environments increasingly different from those they encountered in training. They can fail at generalizing from the values taught and reinforced in their training process to those new situations; and it can be hard for us to be sure how they will act.
And:
We are facing a transition to a world with incredibly intelligent machines, and we need to ensure that transition works out well for humanity. We need to find ways to preserve human agency and enshrine an intrinsic value to being human, in a world where most tasks could be performed by AI.
When I read this sort of thing, it strengthens my belief that the big AI labs are fundamentally different from the ad-based big tech companies in Silicon Valley like Meta. In terms of public backlash, they are all lumped together. It’s that foundation of distrust in big tech that the more significant frustration with AI is built on.
But the technology and vision are different. It's hard to imagine Meta being as thoughtful as Jakub is in the essay linked above. Mark Zuckerberg does have his op-eds, but they usually ring hollow.
With the 40th anniversary re-release of the original Transformers movie coming up, I finally watched the 2007 Transformers. I had skipped it because it looked all wrong. The design doesn’t look like Transformers to me. Also watched Bumblebee, which it much better, just a better feel, better sound.
Nitter and XCancel are back online after being shut down for a couple weeks following a cease and desist letter from X. I'm a little surprised. From the GitHub project page:
Following legal advice, the Nitter project will continue. More details will be announced soon.
Working on some web timeline design tweaks that will go live in the morning. I've been stubbornly insisting on text labels for Reply, Bookmark, etc. where every other network uses icons only. Going to mostly keep that, but the new version looks way better.
Worrisome reporting from Mark Gurman:
Ternus and services chief Eddy Cue want to make even more money from the App Store and figure out ways to raise margins and squeeze additional recurring revenue from the platform. Schiller, on the other hand, seems to believe that such moves will only further irk developers and governments. While there was no internal blowup or anything like that, it’s something he wanted no part of.
John Gruber adds:
I sure hope that’s not true, because that would be awful — making a bad situation worse — and pretty much the exact opposite of what Apple ought to do.
I think I would really like the foldable iPhone. But at over $2k, I can't justify blowing that much money on a phone. It would be a crazy business expense when I could put that money back into Micro.blog hosting or paying people. Other folks may have different priorities, and that's fine! 💰
Pretty thorough article at The Economist about whether AI is creating or destroying jobs:
Add it all up, and The Economist estimates that AI has so far created around 1m new jobs in America. That easily exceeds the roughly 200,000 lay-offs attributed to AI since mid-2023, and appears more than enough to offset weaker hiring in many back-office roles. America’s AI infrastructure splurge has created many of them.
You could argue that construction-related jobs are temporary, but I don't think so. There's going to be demand for years.
I got some great feature and bug fix suggestions on Friday. Over the weekend I've implemented several of them (some shipped already, some need client app updates).
Against my better judgement, considering writing a very streamlined Git app for Mac. No apps are quite right for me anymore. The most important thing now is a big, readable diff view for reviewing AI changes. Commit and push in one step.
Android folks, check out the beta of @gregmorris's Micro Social app for Micro.blog. Google is picky about requiring testers before releasing an app, so even a quick download and poking around helps.
Seems clear from people's fancy demos and my own testing that computer use and spacial reasoning in Astra are big steps forward. It can control Acorn and use selection tools to crop out complex areas, for example, very quickly. Mind-boggling.
Had a dream last night that I went to Best Buy to buy an extra Jaz disk. They didn't have any, but I was able to get one from Amazon that arrived to the store within minutes. Not sure what it means... Maybe I'm worried about backups.
Thinking about the OpenAI industry's PR problem, I'm reminded of Anthropic's ad during the Super Bowl. People said it was clever, but it reinforced the idea that AI companies are just like other big tech: tracking you and selling your data. For next year's ads, we will see a more optimistic vision.
Relocated to work at the Laura Bush library. The view is just as beautiful as I remember when I was visiting 30 libraries in Austin. Can't believe that was 10 years ago.
I was confused for a minute why the coffee shop / bar got so crowded as the morning went on, then noticed that everyone was wearing burnt orange shirts. Hook 'em. 🏈
Developers of popular open source projects are inundated with pull requests. What about a solution somewhere in between a user feature request and a code diff? User writes a feature request, AI checks the project code and turns the idea into a summary that highlights pros/cons based on the codebase.
A couple times over the last year, I've turned off fediverse posting from my blog. This is another one of those times. I may keep fediverse posting disabled until Trump is out of office, when I hope it'll be easier to lower the temperature in online discourse.
Marco Arment blogs about his old neighbors and the divisive nature of our current political environment:
If I learn today that my neighbors have political views that my “side” considers unacceptable and unforgivable, should I be able to have dinner at their house, shovel their snow, or even greet them outside?
Really good post.
I've got access to Astra now and have no idea what to build, as if I suddenly have writer's block. Anyone have a feature request or bug that we've overlooked?
I think my best essays have short titles that really undersell the scope of the blog post. No clickbait, no tl;dr headers to skim through. I hope people read them, but I'm not going to jump through hoops to make that happen.
In a call with reporters announcing GPT-6 Astra, Greg Brockman talked about whether Astra qualifies as AGI:
I do leave it up to the reader to decide for themselves if this qualifies for them. For me personally, I do think we’re there.
As we reach AGI, whether it's with Astra or a future model, I'd like to set some ground rules for myself — beliefs that won't change as technology changes. I don't believe that AI can have consciousness. Ever. AI will be smarter than us, but consciousness and "a soul" can't be created from computation.
When a chatbot says something that gives the impression of having feelings, it's just us projecting human traits onto the AI. My wife and I also named our robot vacuum. That doesn't give it sentience.
AI cannot love. It cannot feel joy, heartache, and despair. AI may be able to say something kind to a friend who is hurting, but it cannot care that the friend is hurting. It's these limits of AI that will reveal what humans are uniquely suited for. What we should focus our lives on.
For every task that AI can do that “replaces” an aspect of our own work, there will be something deeper and more important to work on.
I’m a programmer. In a single year, AI went from being useful in helping me code to now writing better code than I can write myself. And yet I have no shortage of things to work on, to plan, to shape, to ship.
AI might be able to write, but it has no childhood to remember, no person it misses, no mistakes it regrets. It doesn't have experiences to turn into a story. AI might be able to paint, but it can't put its passion into a work of art. It has nothing inside itself that needs to be expressed, nothing that can resonate because it comes from human experience.
Some people fear what AI will obsolete. I welcome it. If an AI can do something better than I can, then maybe that wasn't the part of the work I was supposed to be holding on to.
I realize that this framing is a luxury not everyone has, especially someone who was laid off by a company squeezing higher profits out of automation. Even though my career has been completely upended by AI, I see a clear path forward. For others, that path may be rougher, more obscured, although I hope they find it.
The things that will provide the most meaning in our lives are the things that get us closest to humanity, to love, to heartache, to joy. As intelligence becomes abundant, we can choose to set aside the tasks that are easily replaced, and I think we'll be left with more meaning and a deeper connection with others. When we reach AGI, let's look for what it means that some things won't change.
The GPT-6 Astra announcement yesterday was a bit anticlimactic because most people don't have access to it yet. That's fine, plenty of things to work on with the old-and-busted GPT-5.7 Sol.
Also listening to the interview with Greg Brockman on Stratechery this morning.
I've been working on Micro.blog for about a decade and only now does it feel like it's really starting to get good.
James Talarico is a special candidate. Just watched this video of his rally in Plano. Feeling optimistic seeing that crowd in the Dallas suburbs. 🇺🇸
I don't usually repeat all the little deploy notes from news.micro.blog, but I'm especially happy about fixing video playback in the timeline. It has bugged me for a while that the aspect ratio or dimensions get wonky briefly when clicking play. Much better now!
Catching up on the foldable iPhone rumors ahead of next week's announcement... Kind of bummed it's called Ultra. Fold or Duo could've been great. Maybe at $2k, they needed a more serious name.
I didn't see Apple's Knowledge Navigator video until the early 1990s, a few years after it was produced. I was pretty struck by it. It has taken a few decades, but we're basically there now: OpenAI's Astra intro video.
Bernie Sanders is pretty freaked out about the Hugging Face hack. From a quote on Politico:
Nearly every day, there is a frightening new story about how Big Tech companies are losing control of the technology they are developing, with potentially cataclysmic results.
Uh, cataclysmic? An earthquake is cataclysmic. The hack on Hugging Face, while quite bad, didn't hurt anyone. Working on stronger safety measures is the right response and that appears to be what OpenAI has done.
Fun teaser from OpenAI on X (sigh) about Astra's presumably imminent release. I guess it's going to be called GPT-6. I'm working on Micro.blog improvements today, like always, but have my eyes on the wider social web for this release, so my posts will probably be a bit AI-heavy today. Apologies. 🤖
Part of Kawhi's statement after the NBA investigation into the Clippers. Sort of a non-apology apology:
Integrity and respect for this game are fundamental to who I am. I accept full responsibility for lapses in judgment by people within my inner circle...
I feel like the punishment is very lopsided against the Clippers. No first-round picks for 5 years. 🏀
Firefox adds a global AI opt-out checkbox:
Opting out of upcoming and current AI features on your browser should not require endless navigation through multiple Settings pages. That’s why Firefox offers an AI controls section within its General Settings panel. A single, easily located place where you can block current and future AI features and related pop-ups with the swipe of a toggle.
Kind of similar to Micro.blog's AI checkbox, but the Firefox version is more complicated, allowing control of various AI-related features.
Weird AI ordering system at the Panda Express drive-thru. I’d love to know what model they use. It was a little flaky (“please review the order on this screen…” but the screen showed an ad).
People always say they don’t want AI shoved everywhere. I think that’s partly because it’s used so poorly.
From an article about George W. Bush in The New Yorker:
For Bush, painting is a way to say what he can’t say, to express feelings that include rejection, frustration, and anger, even as he abides by his official discipline of silence.
I couldn’t have been more against George W. Bush in the late 1990s and 2000s. In the Trump world, I’ve become more sympathetic to him — as a human being, at least, if not on policy. He has also become quite a good painter.
Released a new Mac update, version 4.0.8. Really happy with the changes in this one. Lots of little improvements.
One of my favorite things about blogging is when I have to learn something entirely new just to write the post.
Fable 5.1 was released this week. Astra might be coming as soon as tomorrow. Sam Altman, in a long post on X that should really be a blog post:
We have been living with the tension between being excited and anxious about progress for some time, and it is still discordant for us. We know it is much more discordant for other people. And yet, we believe strongly that the world needs to understand where AI is going and how models perform in the real world. More importantly, we believe the world will need aligned AI to manage the future phases of this transition.
In other words, shit is getting real, and the only way out is through. It's not practical to pretend it's not happening or to expect that an elite few people will control the release of capable models, especially as open weight models from China will keep advancing no matter what the frontier labs do.
Sam Altman expanded on this approach of iterative deployment in an interview with Alex Heath. I think Alex did a great job with his questions, covering a range of issues including data centers.
The Information broke a story last night that Astra uses recurrent depth or a so-called looped transformer. The technique effectively runs information through the same block multiple times, with more reasoning happening internally instead of in a chain-of-thought in English that can be inspected by humans.
I wonder if this was a purposeful leak, to get some potentially controversial news out of the way early, and it unexpectedly blew up. I don't know enough about the technical details to judge whether it is worrisome. Jakub Pachocki of OpenAI pushed back on X, though he also wrote that chain-of-thought monitoring is generally becoming more fragile:
OpenAI has worked to preserve and utilize chain-of-thought monitoring since our very first reasoning models. We deeply care about this technique, as it can give us a view into how model alignment generalizes from its training distribution. I do think it is fragile and unfortunately trending in a negative direction, for reasons not contingent on architecture changes that I will write about soon. But there are things we can do to strengthen it, and it's a core goal of our current research program.
Astra does appear to be a major step forward. GPT-5.6 Sol is very good in my experience. I'm pretty excited to see what Astra can do.
The downside of running a tiny business is the "CEO" sometimes doesn't have enough time for the big picture. I was thinking about this yesterday as I focused for hours on Mac development. With AI, code feels free but it's not really free. Everything takes time away from something else.
Deploying a range of optimizations today, across the Micro.blog servers and Mac app. A couple things are dramatically faster. A couple things a tiny bit faster.
Starting to see the gold Tesla Cybercabs in Texas. I don’t plan to ride in one, but to their credit it sounds like the cameras might be better than the ones in a normal Tesla. I just don’t get stubbornly ignoring LiDAR when the technology is right there.
Of course Waymo agrees. From a Waymo blog post:
Cameras are incredible, but they aren't enough.
Cybercabs can probably drive a little better than the average human, but not much better.
Kind of love the all-lowercase "hello" as John Ternus's first X post. Maybe we're in for a new, more informal style. Or maybe I'm reading too much into it. Still holding out hope for a CEO personal blog one day, to offset all the opaque corporate-y communication.
One way to read the OpenAI vs. Anthropic rivalry — and Claude Code's success — is that while OpenAI was off in the weeds playing with Sora, Anthropic was training Mythos. OpenAI spent 2026 catching up. Reading about Astra rumors, I expect OpenAI to have a significant advantage through the fall. 🤖