@schiste

Currently building in the shadows, all day long - Former chair of @Wikimedia & @Wikimedia_Fr - he/him

 Paris
schiste.substack.com/
Joined March 2007
  • Tweets36.6K
  • Following1.6K
  • Followers2.9K
837Media
Tiens marrant, une manifestation sausage pendant un couvre feu. Ils sont à deux doigts de sortir les flambeaux tellement ils sont pepouzes.
Impact
@ImpactMediaFR8h
🚨🇫🇷 ALERTE INFO ! Alors qu’un COUVRE-FEU EST INSTAURÉ À LYON, des militants d’extrême droite défilent dans les rues de la ville en chantant LA MARSEILLAISE. (témoins)
1
1
313
Il y a 20 ans, le web et le mood c'était plutôt ça : youtube.com/watch?v=zlfKdbWw… et je fais mon vieux con mais pour le coup, c'était vraiment mieux avant.

Where the Hell is Matt? 2008

www.wherethehellismatt.com14 months in the making, 42 countries, ...

youtube.com
1
1
2
434
Christophe Henner retweeted
Steeve Morin
@steeveOct 16
zml/llmd is now faster than vLLM on DeepSeek-V4.1-Flash on 4xGB300. The remaining 30 tok/s are already accounted for and will be gone very soon. And most of those optimisations generalize to other models. Little by little.
6
4
41
2,287
Christophe Henner retweeted
Quan Nguyen
@stablequanOct 14
Ummm guys, I took @googlegemma Embedding 2 model and train it on 3,000hrs of Vietnamese audio. IT REACHED SOTA. All it took is 2 hours on 8xH100
62
82
11
1,287
68,822
Bientôt prêt de chez vous, les botnets de SLM autoreproducteurs : anthropic.com/research/self-… vous trouviez les attaques cyber violentes, le jour où des conneries comme ça tournent réellement ça va pas être la fête.

AI models may be capable of self-replicating without access to a datacenter

Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.

anthropic.com
2
113
Si on se comportait avec l’Europe comme le fait l’Allemagne, on serai dans une autre situation. Je suis un federaliste acharné mais l’Allemagne fait tellement de l’anti jeu…
1
110
Après le blocage de la préférence européenne de la commande publique, le maintient du thermique dans les voitures, le scaf, ESSI et maintenant MGCS (char du futur), l'Allemagne commence à empiler les actions anti construction européenne...
1
150
Hi @ecosia and @Christian_KrolI the fact that Ecosia relies on search results from US providers is, in my view, not truly sovereign. Talking about sovereignty while depending on US tech and moving to depend on Chinese tech is a very interesting, though not very rational, stance.
5
84
And the first sentence is paraphrasing the founder in the Politico article : politico.eu/article/germany-…

German search engine ditches Mistral, bets on Chinese open-source AI

“We are disappointed with the quality of Mistral,” Ecosia founder and CEO told POLITICO.

politico.eu
6
52
Germany is proving to be the least pro European country over and over again.
16
2027 va être l'année de la casse sociale de la presse avec des chiffres comme ça.
Emmanuel Parody 🇪🇺 🇺🇦
@eparodyOct 8
Le bain de sang du trafic des sites d'information : les sites d’info perdent 23% de visites, les applis résistent, selon l’ACPM fr.themedialeader.com/ai-ove…
1
1
179
Christophe Henner retweeted
Christophe Henner@schisteOct 7
Like OpenAI models, Haiku is cheaper on short context, Add under modelSettings in ~/.claude/settings.json json { "modelSettings": { "claude-haiku-5-5": { "autoCompactWindow": 100000 } } } You will save a shitload of quota without doing anything.
ClaudeDevs
@ClaudeDevsOct 7
Replying to @ClaudeDevs
For prompts under 100K tokens, it's $0.10 / $0.50 per million tokens with cache reads at $0.01. Above 100K tokens, $0.50 / $2.50 per million, with cache reads at $0.05.
1
2
1
190
Comme les modèles OpenAI, Haiku est moins cher sur les petits contextes, ajoutez aux modelSettings ~/.claude/settings.json { "modelSettings": { "claude-haiku-5-5": { "autoCompactWindow": 100000 } } } Des économies faciles sans trop réfléchir :)
ClaudeDevs
@ClaudeDevsOct 7
Replying to @ClaudeDevs
For prompts under 100K tokens, it's $0.10 / $0.50 per million tokens with cache reads at $0.01. Above 100K tokens, $0.50 / $2.50 per million, with cache reads at $0.05.
1
108
Like OpenAI models, Haiku is cheaper on short context, Add under modelSettings in ~/.claude/settings.json json { "modelSettings": { "claude-haiku-5-5": { "autoCompactWindow": 100000 } } } You will save a shitload of quota without doing anything.
ClaudeDevs
@ClaudeDevsOct 7
Replying to @ClaudeDevs
For prompts under 100K tokens, it's $0.10 / $0.50 per million tokens with cache reads at $0.01. Above 100K tokens, $0.50 / $2.50 per million, with cache reads at $0.05.
1
2
1
190
Si vous aussi vous aimez tester les trucs de Pentest Automatisés, j'ai fait une liste qui se maintient à jour toute seule : github.com/schiste/automated…

GitHub - schiste/automated-pentest-list: Curated directory of public GitHub repositories for...

Curated directory of public GitHub repositories for automated penetration testing, autonomous pentesting, and AI pentest tools. Compare scope, execution model, license metadata, and maintenance. - ...

github.com
87
2026, LLMs have seen so many .env that keys are safe because there are too many. A bit like French citizen private data being safe because it already leaked 12 times this month.
2
6
275
Here is the self-updating Pentest Automation list/folder/repo : github.com/schiste/automated… There are a lot of projects going on, and the field is interesting but it's a nightmare to find them and see which are really active projects versus week-end projetct.

automated-pentest-list/catalog/directory.md at main · schiste/automated-pentest-list

Curated directory of public GitHub repositories for automated penetration testing, autonomous pentesting, and AI pentest tools. Compare scope, execution model, license metadata, and maintenance. - ...

github.com
1
80
And I've been testing some, but didn't keep track etc... So I automated discovery, enrichment and tracking and make it possible for people to leave opinions, reviews, etc in github (submit PRs).
37
I don't think Texas Instrument realize the harm it does to kids. Why would they learn to count when they can use the ABLE calculator and let the machine do it for them. #tweetlikeits1976
71
Oh mais ça ne serait pas le Space Bunny qui tourne actuellement ? Ca expliquerait la difficulté à identifier le modèle. Et vu la communication sur ses performances cyber ça expliquerait aussi pourquoi Space Bunny fait ses audits de manière très différente des autres.
Mistral AI
@MistralAIOct 6
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
1
1
231
Si c'est le cas, ça serait top que Mistral confirme ça permettrait de partager des expériences de premières mais des 2 dernières semaines ^^
1
1
77
Et la fin de l'usage free de Space Bunny est le jour du lancement de Mistral Large 4. @MistralDevs @MistralAI Space Bunny c'était Le Chonk?
1
107
Load more