On finding joy in slow, imperfect code in an age of copilots, agents, and chatbots. 1.2k words, 5 min reading time.

Three weeks ago, I wrote a the post Rescuing Scholia: will we make it in time?, where I sketched a future without Scholia. Scholia, started almost 10 years ago and I think it is worth keeping around longer.

There’s an old parable about a drunk searching for his keys under a streetlamp. A passerby stops to help, and after some fruitless searching asks, “Are you sure you lost them here?” The drunk replies, “No, I lost them in the alley. But the light’s better here.” I think about this story a lot when I read AI predictions.

Starting this week, the Rogue Scholar science blog archive has consolidated authentication into a single Identity and Access Management (IAM) service, powered by a self-hosted Keycloak instance at https://auth.front-matter.de.

AI creates an education challenge, not just a job crisis. We haven't built systems to help people to continue learning and connect them to new opportunities.

The annual “Molecules of the Year” selections are available for the year 2025.

This is what Struck wrote in 2018 in a contribution to the 2018 IEEE 14th International Conference on e-Science (e-Science) (doi:10.1109/eScience.2018.00016). I very much agree with this, and the notion is gaining ground in the academic community.

Green and gold of autumn My annual selection of favourites from the photographs I took in the past year is now available on Flickr. Do people still use Flickr? I have broken my usual rule of not including family photos because of the very exceptional and very happy occurrences of both of our daughters’ weddings this year. It would have felt wrong somehow to omit pictures that captured the sheer joy of these events.

If an AI produces something useful it's because of your own skill in model choice, prompting, and steering. If not, it's because the model is a useless lying machine that can't follow directions.

TipHey, I’m writing a book about this! I’m actually writing a book about this stuff. It turns out there isn’t a lot of literature on how to do post-training at the level too big for single-GPU laptop-sized hobby projects and requiring enterprise reliability on one hand, but not quite at the scale of multi-team distributed post-training you’d get in foundation labs.

Since I last wrote, I have had a few more thoughts on Samuel Moore’s book. Again, these are not necessarily things that he does not discuss or things that he should have discussed. They are merely thoughts that occurred to me in response to reading his work.