331 private links
Additionally the "fix" fixes more than 40 separate issues in a single mega-commit touching 68 files which like. This is really beyond "you are trusting the slop machine too much". This is "the moment the slop machine got in your hands, a switch flipped in your head and you were like 'I never have to follow any engineering standards, any at all, ever again'". To repeat once again: "AI" is not a technology, it is a permission structure.
The good engineering practices are ignored by AI and the people seems to not care about it.
Des réflexions sur la stabilité de quelque chose développé par IA. L'exemple est celui de ArtCraft: https://shaarli.lyokolux.space/shaare/aaHWxg)
The clones are not feature complete.
An AGENTS.md for students
OpenAI submissions on math related topics
Now we have generative AI, which can also do this work. To a person who doesn’t care about the craft, some generated documentation or design elements are exactly the same as those handed over by a writer or designer. Someone, or something, else has done the work. However, machines don’t care. [...] Design systems and editorial style guides need people who care. They need people who care about the small details, who obsess over consistency. They need people who are willing to push back, and who are happy to say no to the endless requests to ignore the guidelines just this one time.
he problem for us is that it’s very difficult to demonstrate the impact of it until it’s gone
1) Quand un logiciel est opensource, vous pouvez accéder à son code source, le lire, le comprendre et le modifier (ces deux mots sont importants). Un modèle de LLM, ce sont des milliards de paramètres en virgule flottante que personne ne peut comprendre. Personne n'est capable de dire, par exemple, comment est encodé le concept de "pomme" dans un tel modèle. Et il est donc encore moins possible de le modifier.
2) Une LLM "opensource", c'est comme si je vous donnais un programme binaire déjà compilé (un .exe) en vous disant : "C'est opensource ! Il contient toutes les instructions en langage machine !" [et le reverse engineering est impossible] (À ce titre, Windows est opensource, hein.) Il serait plus juste de parler de modèles gratuits, mais c'est tout.
3) En plus de ne pas pouvoir les inspecter, vous n'avez aucune idée des données qui ont servi à entraîner ces modèles "opensource". Vous ne savez donc même pas ce qu'ils contiennent.
Comment arriver à générer autant d'ici 4 ans?
·
A carefully written piece
Josh Comeau notes that with LLMs, "there’s no incentive for people to make high-quality free content."
[Rachel Andrew] reports that GenAI has broken the relationship between subject matter experts – the authors of the original draft – and editors who proof-read, fact-check and polish the article. LLMs output needs to be reviewed by people who deeply care about the end result.
Today, we often hear the word “democratize” from GenAI promoters, but it means the exact opposite. AI centralizes power and commodifies knowledge, instead of teaching skills and deep understanding.
RigSpark scores your hardware and gives every model a yes, slow or no verdict with estimated tok/s before you download. Then it verifies, serves and chats with your pick.
The source code is on Github.
des skills pour agents IA que Claude Code, Codex, Cursor ou Mistral Vibe peuvent consulter pour endosser le rôle d'expert-comptable, de contrôleur fiscal, de commissaire aux comptes, de fiscaliste, de notaire ou encore de syndic de copro.
Cela peut être utile.