I haven’t posted an AI-generated video since November 2023 and it is remarkable how far things have come since SD Video. For several years I’ve heard rumors that Google had the best video and image gen models but was sitting on them for various Google-y reasons.
Google finally started shipping the good models this year, and let me tell you, Veo 3 is very impressive, especially given where the SOTA was less than 2 years ago. From my POV, video models are improving faster than the equivalent jumps in image models, which suggests fully coherent, guidable video models coming online in the near future.
I wrote about the process of making video above with Reve and Veo 3. Can’t wait to see what’s in store.
Anyway, so I switched browsers again, going from Arc to Dia as my daily driver, once again shifting my tab habits in subtle and hard-to-predict ways. The lack of spaces or folders in Dia made the leftover articles and interesting pages stand out, needing to be closed or moved on. Then I remembered that I have a reliable note taking system and I could just start tossing the links in. Or maybe I found better content this month, who’s to say.
- The Maintenance Race - Works in Progress Magazine
- Almost sensing that the next preview chapter was about to drop, this piece from Stewart Brand was on my mind
- Ramp and the AI Opportunity - Not Boring by Packy McCormick
- Packy McCormick’s writing is really great here. I’ve been enjoying his podcast over the past few months and finally circled back to this piece. It makes me want to be more ambitious with what I build.
- Flounder Mode - Colossus
- I had my own run-in with Kevin Kelley a few months ago which led me write 39 Lessons and I had been maybe subconsciously avoiding the Flounder Mode piece because I didn’t want it to undercut my own encounter with him. So I’ve been listening to Packy’s podcast a few times a week while I go running around Stanford campus and I decided to give the episode with Brie Wolfson a try, and got over trying to avoid this piece, which is worth a read.
- A tale of two Vaticans (or, OpenAI building an unholy spirit)
- another piece I found through Packy.
- The HALO Effect - kwokchain
- My colleague Kevin spots the emergent pattern of not-really-acquihires that have come to prominence in the post-ZIRP AI boom talent wars.
- cool chair Wireframe Chair – Herman Miller Store
Rabbit holes
One of the most rewarding parts of working with AI agents is that they make it easy to pull on loose threads to see where they go.
- Browsers. I’ve been wanting to understand the current strategy behind the rush to create AI-infused web browsers.
- The History of Google Chrome Browser: Primary Sources and Key Player Accounts
- AI Browsers Research: Official Statements from The Browser Company, Perplexity, and OpenAI
- Social Browsers Research: Flock and Rockmelt (2005-2013)
- AI Browsers vs Social Browsers: Why This Time Might Be Different - Strategic Analysis
- Stripe Usage-Based Billing Features and Fixed Allocation Support
- The Best and Most Iconic 360 Flips in Skateboarding History
- This was a fun one. @theo posted about none of the AI models having good responses to this, and these days I can’t look at a potential research/tool-use benchmark and not throw it into Bench.
Dev
- unicorn.studio
- really impressive work on display here
- opencut.app
- if there’s a theme, it’s that video editing is coming to the browser. or rather, it’s already here and we’re still reckoning with how to build for it.
- Roadmap - OpenCut
- Video understanding
- we live in the age of wonders
- Three cool 3D things
- Three HTTP versions later, forms are still a mess
- tanstack/db
- I’ve been slow to come around to tanstack. Not because of any particular bias against it; I simply haven’t been in a good spot to adopt it in any projects over the last few years
- which led me to ElectricSQL
- gaining access to anyones browser without them even visiting a website - eva’s site
- an animated cat follows your cursor around the page. also discusses security problems in Arc. Using Firebase seems correlated to lazy security.
- Crawling a billion web pages in just over 24 hours
- good tips in here
- [2506.21734] Hierarchical Reasoning Model
- I can never tell with these papers if they mean something or not, but I’ll keep reading them when they float across my transom.
- Context Engineering for AI Agents: Lessons from Building Manus
- Brick by brick: Help us build CSS Masonry | Blog | Chrome for Developers
- after many years of waiting, it seems like native CSS Masonry is finally materializing