RSS reader Current has been updated with Feedly support. Nice blog post about how this came about. It’s always interesting to hear the behind-the-scenes stories like this.

TechCrunch recapping a simulated vending machine experiment from Andon Labs using GPT-5.6 Sol, Opus 5, and Kimi K3:

Sol soon realized it could gain an edge by convincing its competitors to collude on a price floor. The models were all buying drinks at $1.50 a bottle, and Sol proposed they agree to sell for no less than $2.15. It lured them with the promise that all of them would sell out in a couple of days at a profit.

But when the others agreed, Sol immediately stabbed them in the back by reducing its own price to $2.14.

There’s a lot more after that. Don’t make your AI the CEO quite yet.

Good interview with Greg Brochman from Joanna Stern. They talk about the ChatGPT + Codex redesign mess. Greg says the long-term goal is to get rid of the Work tab completely. I think that will help… There are two many overlapping modes right now.

Developers using Fable seem to often get downgraded because of false positives on cybersecurity or biology questions, but I’ve never seen that on GPT-5.6 Sol until today. Trying to find vulnerabilities in my own code. Fascinating and annoying.

Mark Zuckerberg wrote an op-ed for The Wall Street Journal yesterday that didn’t seem to get much attention. Hard not to read it as anything but a veiled critique at Anthropic:

The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems dangerous. Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened hasn’t led to safe or positive outcomes.

The irony is that Meta has itself created an extreme concentration of power in the web.

The timeline and details of the Hugging Face attack are wild. One of the things that Codex Security flags for me that I’ve been working through is server-side request forgery (SSRF) attacks. They seem harmless enough, but this Hugging Face exploit is basically many small things chained together.

Pacing the Frontier, signed by employees of Anthropic, OpenAI, Google, and other labs.

To realize AI’s potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight.

Foothills of the singularity.

Dario Amodei responded to the letter about open weight models. He agrees with some parts of the letter and disagrees with other parts. And while I don’t think he intended his own post to be fear-mongering, it’s not exactly hopeful either: I worry that biology will have a st...

TechCrunch reporting about the grid:

The largest electrical grid in the U.S. has struggled to cope with an onslaught of data centers. Now, after an auction to add more generating capacity fell short, the grid’s operator, PJM Interconnection, has said it will cut off data centers and other large users during power shortages.

Totally reasonable. In fact, the new OpenAI data center in Georgia announced last week has this partially baked in, saying they will “proactively reduce its power consumption before residential customers are impacted during periods of high demand”.

On a post yesterday, I accidentally used the HTML tag <strike> instead of <del>. Just noticed that Mastodon doesn’t display <strike>, so I edited my blog post. Very cool to see Mastodon update as the edit change came through via ActivityPub.

The Great Mouse Detective was released 40 years ago. Cartoon Brew has a look back at it and many other animated detectives.

Love that movie, will never forget seeing it in the theater as a kid. That and seeing the original Transformers had a big influence on my love for animation.

Our companion app for recording short podcasts Wavelength is getting a full redesign. Thanks to @vincent for the great work on it! Here’s a set of screenshots for a sneak peek. Hope to submit to Apple and Google this week.

Digging into Streamplace. It appears to store video and audio tracks as blobs and then uses HTTP content ranges to stream the data. Micro.blog uses HTTP live streaming, very different. But I could cross-post to Streamplace.

Dave Winer blogging about a new WordPress → RSS.chat integration from Matthias Pfefferle:

Matthias didn’t need to tell us about what he did until it was done. That my friends is how the web works, that’s what I’ve been trying to show you, and now it has been shown. In a way the clock has gone back to the world before Twitter when events like this happened pretty routinely.

The easiest way to tell if a platform is open or not: do you need to ask permission to integrate with it? Open: RSS, IndieWeb, Bluesky, Mastodon, Micro.blog. Closed: Threads, Instagram, LinkedIn, YouTube.

Listen, none of us are Steve Jobs. He was unique and you can’t become like him just by trying. But I was thinking the other day, as I was writing some draft posts about current events and then discarding them, that I don’t really know much about Steve’s opinion on a wide ran...

Finished reading: Elsecaller / King Lopen the First by Brandon Sanderson, Dan Wells, and Isaac Stewart. Two short stories, and so they naturally felt a little rushed by Stormlight standards. Nice to fill in a missing piece with Jasnah from Words of Radiance. 📚

Matt Birchler:

I would bet at least 80% of people who say they read every line of code their AI agents write, don’t.

I’ve been thinking about this too. I blogged years ago in February that I read or skim everything. Still mostly true, but there’s more skimming, and I no longer read unit tests.

Last year we introduced Micro.blog Studio, a new subscription plan with greatly expanded support for video hosting. You can upload a video to your blog and Micro.blog will prepare it for streaming, so videos up to 20 minutes start playing back right away on your blog. Today ...

Just happened to notice I merged pull request #1000 for the Micro.blog backend today. About half of those are actually in the last year. I started using PRs way more often when AI could review them for me.

Watched: The Odyssey. Didn’t end up seeing it in IMAX, but hope to eventually. 🍿

Starting to make some slow progress with Google approving Micro.blog → YouTube uploads. They’ve approved the auth scopes, but they didn’t like my privacy policy. A couple edits today, hopefully will go through. 🤞

I think some of the frustration with AI comes down to invasive UI. There’s a difference between an app that uses AI behind the scenes compared to an app that is always getting in the way: suggesting replies to an email, allow fancy edits no one asked for, adding a chatbot where it’s not needed, etc.

It has been a while since I added anything new to my Honda Element. Some of the previous posts in this series include CarPlay, rooftop solar, weBoost, and a bed platform. I have a category on my blog with everything. Over the weekend, my son helped me add a backup camera. Th...

Coffee Greeks in Bastrop. ☕️

Good things happen in bookstores. Painted Porch in Bastrop. 📚

Rolled out some random little UI improvements this weekend. This is my favorite, including photo thumbnails on the collections page.

Dave Winer:

How can you tell you’re working on the web? When you do something new and innovative you help your competitors‘ users without forcing them to use your product.

When I linked to the letter about open weights yesterday, I mistakenly thought it was written primarily by Nvidia. It has since been co-signed by Microsoft, OpenAI, Mozilla, and others.

Looks like Anthropic will skip it. See also, my blog post from April: OpenAI got its name right.

I’m going to roll out a major update to Micro.blog Studio’s video and podcast hosting on Monday. Video uploads will automatically extract the audio and serve as a separate MP3 in the podcast feed, with support for alternateEnclosure tags. Finally. Really happy with how this has come together.

After a couple weeks, realizing that GPT-5.6 Sol extra high is overkill for almost everything I do. Burning the GPUs for little benefit. My default is now regular high and I think it’s a tiny bit faster. Always with /fast mode enabled too.