It's been a couple of years since GPT-4 powered Bing, but with the various Deep Research products and now o3/o4-mini I'm ready to say that AI assisted search-based research actually works now https://simonwillison.net/2025/Apr/21/ai-assisted-search/
"Taking care of your own problem-solving mind no matter what the AI can and can't do: a user manual" ?!!!???
We got our AMD Turin sled booting! And --in classic @oxidecomputer fashion -- we're going to regale with tales of the adventure: join @ahl and me on Monday at 5p Pacific when we will have the engineering team on hand to talk about bringing up Cosmo!
https://discord.gg/QrcKGTTPrF?event=1362938379364466941
If you want the backstory, check out our earlier tales from the bringup lab:
https://oxide-and-friends.transistor.fm/episodes/tales-from-the-bringup-lab-2021-12-06
https://oxide-and-friends.transistor.fm/episodes/more-tales-from-the-bringup-lab
And learn ,more about Turin:
https://oxide-and-friends.transistor.fm/episodes/unshrouding-turin-or-benvenuto-a-torino
@ren my theory is that billionaires mostly have free associative and reactive thoughts about what they feel and then there is a whole apparatus organized around them to translate their amorphously bad feelings into plans that make essentially no sense, but do make sense if the goal is to make the billionaire feel good
That piece of writing I was talking about finishing that I was equally proud of and scared to post because I love it? Here it is.
I am starting a new project, it is A Newsletter, and it is About Tech, but it is a very different tech newsletter.
Welcome to Fight for the Human. Sign up to start setting your compass toward rehumanization.
TextArena as a Benchmark:
Scaling and reasoning led to significant progress on frontier benchmarks, saturating many of them. We believe that relative measures of success present a more sustainable future for benchmarking, and thus created TextArena.
In TextArena models can compete against other models and humans in over 70 text-based games. Results are updated in real-time and shown on https://textarena.ai/leaderboard. (You can play against the models here: https://textarena.ai)
Thanks to @sbidoul , pip 25.1 will have a `pip lock` command that uses `pylock.toml`!
https://github.com/pypa/pip/pull/13213
pip-tools has started looking at if they can leverage it.
https://github.com/jazzband/pip-tools/issues/2124
pip-audit has support in 2.9.0 .
https://pypi.org/project/pip-audit/
@frostming has a PR for 'packaging' to add the required marker support (I assume for PDM support). It's getting really close to being merged.
https://github.com/pypa/packaging/pull/888
I'm a bit relieved there's uptake of pylock.toml already!
... and the all-important "Generate an SVG of a pelican riding a bicycle", for o3 and o4-mini - I had them both generate their own alt text for the image they had generated too, using their vision input
Une discussion que j’ai eu avec ma prof en TP de stats en python, à l’instant.
— Elle : Tu sais il vaux mieux donner des noms courts au variables, écrire "matrice_correlation" c’est long on peut juste écrire "cor"
— Moi : Ouais mais après on comprends rien, c’est con. D’ailleurs j’allais demander mais ça veut dire quoi "p" dans ce bout de code ?
— Oh bah ça doit être défini plus haut… *regarde*… ah non… ah je crois que c’est… *réfléchis un peu plus*… ah non ça serait pas logique… *regarde ses notes persos*… ah oui en fait dans mes notes j’avais appelé cette variable "d", ça contient le nombre de colonnes de la matrice.
Okay let's get the videoroll from AlgoRave going!
I'll be posting links to each artist involved in the post, or pinging if I can reach them here on fediverse
m-onz's performance felt like a mad scientist experiment, and they had visuals glitching out during their set... but it was stuttering audio so rhythmically, it might as well be on purpose! Enjoyed every bit of music and all craziness going on!
https://fakedac.net
It's been 2.5 years with little progress finding mitigations for prompt injection attacks LLM apps... but that may finally have changed!
Google DeepMind published a paper describing CaMeL, an ingenious system that could, maybe, lead to secure digital assistants https://simonwillison.net/2025/Apr/11/camel/
code / data wrangler in Switzerland.
Recovering reply guy. Posts random photos once in a while.