348 private links
We ended up deciding: what the heck, we might as well meet the market demand. So we put together a bespoke ASCII tab importer (which was near the bottom of my “Software I expected to write in 2025” list). And we changed the UI copy in our scanning system to tell people about that feature.
I am not sure it's a market demand, but only a ChatGPT hallucination.
the many ways AI is making humans less productive #1:
- I have to prove I am human when visiting website
- websites are generally slower because they need to check if my browser is an AI bot
- I spent time every week now, restarting services affected by AI scraping bots
And I am not even using AI
An interesting project: let the AI squash and reword the commits
MCP:
- Turns M ⨯ N integration problem by turning it into an M + N integration problem.
- Decouples AI client applications from AI tools and workflows for a platform.
That's their reason to exists
The server exposes: Tools that the LLM can call, prompts, resources exposed via URIs and sampling (allows servers to request LLM complections on the client application
An MCP server is another tool in the toolbox.
These fake video can still be spotted on few details, but they start to feel real
Cela ne sent pas bon pour les futurs investissements déraisonnable.
Comme pour toute bulle d'investissement, seule les applications éprouvées continueront d'être financées.
The books were once criticized similarly to AI now.
Saying no to AI is a luxury too. You have to be able to.
AI is not good now: https://github.com/dotnet/runtime/pull/115733
Thus, the general expectation is that AI implies, at the very least, software that consistently and reliably outperforms a human expert at any task in any given field it claims to be proficient in.
The paper in question: https://aclanthology.org/2023.findings-acl.426.pdf
Gzip output could be used as input for LLMs. They won't be "Large" in this case and this is a perfect fit for less ressource consumption.
(via https://nicolas-delsaux.hd.free.fr/Shaarli/shaare/kyvTdw)
L'organisation non-gouvernementale Noyb travaille bien à faire respecter le RGPD.
ce combat porte essentiellement sur la question de savoir s'il faut demander le consentement des gens ou simplement prendre leurs données sans leur consentement
Meta a commenté:
L'entreprise a ajouté : « les recours en série de noyb font partie d'une tentative d'une minorité bruyante de groupes activistes de retarder l'innovation en matière d'IA dans l'UE, ce qui nuit en fin de compte aux consommateurs et aux entreprises qui pourraient bénéficier de ces technologies de pointe ».
Les LLM ne sont pas assez efficace sur des tâches spécifiques et consomment trop de ressources. En ce moment, il y a aussi le retour d'autres types d'algorithmes. Ces algorithmes peuvent être associés mais n'ont pas besoin de LLM.
En revanche, les SLM et les réseaux de neurones vont continuer à être utilisé, puisqu'ils ont des utilités spécifiques.