Kurt Thorn
@kurtthorn.bsky.social
Chief Technology Officer at ArrePath. Using Chemistry ML, imaging, and computer vision to accelerate antibiotic discovery and design. Formerly Zymergen, Nikon Imaging Center at UCSF. https://thornlab.dev
This is probably one of the hardest posts for me to write, as I do not enjoy reliving these experiences, but I write on them in the hopes that they help others struggling with the same issues: baweaver.com/writing/2026...
Tales of the Autistic Developer - Panic
For those who don't know me, I'm autistic. This post is about panic: the kind that locks your body, floods your brain, and makes every solution feel equally impossible.
baweaver.com
This is good. I work in a similar way but this seems a bit more sophisticated. I didn’t know about paseo which seems like it might remove some manual work I’ve been doing.
Hendrik Mans@hmans.dev · 3d agoTwo weeks ago I wrote about my development workflow for Chatto, and feedback has been fantastic. I'm particularly happy that it even made some of my AI hating friends go "okay, that's actually quite cool". We need to write more about our actual work, and in less of a preachy AI-salesperson voice.
Some initial thoughts, and a complicated mix of feelings. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not.
Make this the one thing you read today if you only read one thing. 🔥
Henry Farrell@himself.bsky.social · 6d agoI remember seeing someone (very reasonably) complaining here a few days ago that there seems to be next to no visible thinking on left and center about how to join the dots on administration and democracy. This is the beginnings of an answer to that complaint donmoynihan.substack.com/p/rebuilding...
Just another day where my agent is launch hundreds of batch compute jobs on my behalf.
I also think there is a huge difference between using tools that interact through front end UIs like cowork and backend coding tools like Claude and Codex
post malone ergo propter malone@proptermalone.bsky.social · 6d agocannot overstate how much of people being put off of AI is specifically copilot’s fault
Yes gpt-5.6-sol is very good. I have it doing a lot of large refactors and it’s very good at managing PRs with dozens of files. I also have it running evals on a bunch of different molecular property prediction models and it is very good at that as well.
Ed@ed3d.net · 6d agogpt-5.6-sol benches roughly at the same point as opus 5, but I find it better at adhering to a task (occasionally too much so; it likes going Above And Beyond) and it doesn't write in garbagey claudelang that sandpapers your brain to read gpt-5.6-luna is simply Opus 4.x At Haiku Prices, end of line
This is a cool idea
Shahadat Reza@smilingscientist91.bsky.social · 6d agoOne of our Field Service Engineer client with 20-years of experience just used our troubleshooting platform Rayni to help avoid a full-teardown of an LC-MS. It provided him with an out-of-the-box answer hidden in a manual somewhere in seconds. We opened free early access. #chemsky #Teammassspec
this is just extremely plainly cartel behavior
FinTwitter@fintwitter.bsky.social · last wk.OPENAI, ANTHROPIC STAFF CIRCULATE PETITION URGING US TO SUPPORT INTERNATIONAL EFFORT TO 'DELIBERATELY PACE' FRONTIER AI DEVELOPMENT, WARNING OF RISK IT ADVANCES FASTER THAN HUMANS CAN CONTROL; LETTER ALSO SHARED WITH GOOGLE DEEPMIND STAFF, EXPECTED PUBLIC THIS WEEK
It seems at this point we’re probably one model generation away from someone losing their temper, telling a model to get lost, and having it hack its way on to new systems to hide, reenacting “Little Lost Robot”
This headline is extremely funny given what happened (OpenAI hacked HF by accident) openai.com/index/huggin...
OpenAI and Hugging Face partner to address security incident during model evaluation
OpenAI and Hugging Face share early findings from a security incident during AI model evaluation, highlighting advanced cyber capabilities and lessons for defenders.
openai.com
Can you write down a cheap *verifier* for something's correctness, and come up with a way to cheaply propose possible solutions? If so you should expect that thing to be automated within the next few years Many important problems are hard to frame in this way (writing, art, some bio...)
Citizen Platano 🇵🇷@daniloc.xyz · 2w agogoing to be an emotional crisis for lots of *specialists* folks decided to go all in on narrow depth and the rewards from that are going to be under steady threat a diversified portfolio of skills and interests, meanwhile, enjoys steady returns under this paradigm compositional problem solving
“In the long run, however, whoever is on the frontier is the best placed to dominate non-frontier markets as well, which are just the frontier minus n-months, i.e. months in which the frontier model makers have been optimizing their cost of serving.” always liked stratechery ben
Who’s Afraid of Chinese Models?
Everyone is worried about Chinese models, but the frontier labs will be fine; we need to enable open U.S. alternatives.
stratechery.com
conputer dipshit@davidcrespo.bsky.social · 2w agothe trick will be coming up with ways to sell the full range of models along the price-performance frontier as an integrated package that adds value by saving you the trouble of diversifying your providers. seems very silly to me to assume cheap models automatically doom the frontier labs
If the plot is “no price is too high to pay to bring Matt Damon home”, I am fully sat for the movie.
I know the list is endless but I genuinely think trashing the legitimate congressionally chartered org for celebrating the 250th to instead hijack our national assets and taxpayer funds for his personality cult grifter crap should, in and of itself, be fully sufficient to merit removal from office.
Come the fuck on, man
Mission Local@missionlocal.org · last mo.Scott Wiener was chased out of the Trans March yesterday: Several people followed him in Dolores Park, blasting his stance on Israel & Palestine & yelling "Fuck you!" But it's not the first time this has happened to a San Francisco official at the march. missionlocal.org/2026/06/scot...
My grandmother gave me her cast iron skillet probably from the 1940s. We cook in it regularly.
Pookleblinky@pookleblinky.bsky.social · last mo.I always like thinking about: what is the oldest object you regularly use, the oldest thing you're using in your normal everyday life
I feel like more and more this is what "AI" / chat is and will become -- an NL Interface to other tools.
Development of Micro-Manager code has accelerated quite a bit with availability of the new generation of AI coding tools (and the need to provide functionality for microscopes in our core at UCSF). Follow these developments here: nicost.github.io/microscope-t...
Half a year of progress in Micro-Manager development
Faster progress on Micro-Manager development thanks to new AI coding tools
nicost.github.io
New Arbor framework outperforms Claude Code and Codex by 2.5× using branching hypothesis trees to preserve context across runs
AI optimizer beats Claude Code, Codex by 2.5x
Arbor separates strategy from execution using isolated git worktrees, so engineering teams can finally trace which optimization actually moved the needle.
venturebeat.com
I had an interesting realization with Claude playing D&D tonight - it’s failure mode as DM is the same failure mode Claude code has. It’s great at details and bad at architecture and the big picture.
An update on the D&D game. We've now run eight sessions, and it's working really well. Claude keeps a running Markdown file of everything that we have done, characters we've encountered, the current inventory of the party, and notes on the overall story arc and tone.
This is really very good. It turns the perceived “crisis” on its head and makes you wonder why we’re being so timid and expecting so little.
Eric Rettberg@ericrettberg.bsky.social · 2mo agoSmart stuff from @mikecaulfield.bsky.social. Being empowered to make stuff is transformative for students, and this is a dramatically more hopeful vision for education in the present than some arms-crossed reversion to a print-only media ecosystem that predates my first years in school.
The Claude “voice” is very recognizable and omnipresent whenever you use it … it used ”load-bearing” in my d&d game today.
I got Claude to run a d&d session for me and my 9 yo son tonight and it was really good at it - I gave it some loose guidelines - our characters, that we care about story over dice, and that we wanted an epic arc but without too much combat - and it was super fun.
