Replying to @mattturck
As an ashamed french I think the mockery will help a lot. We need the crazy reaction to go up to their ears
Paul Masurel 🦀 retweeted
Coding agents made AI building fast. Now the bottleneck is understanding what got built.
@dataiku Headless lets developers build data pipelines, ML models, and custom agents from Claude Code or Codex. The output is visual, not 2k lines of code.
tinyurl.com/2f2zy3dc
#DataikuHeadless
Paul Masurel 🦀 retweeted
l’annonce d’openAI sur la résolution de 3 verrous scientifiques majeurs marque un tournant brutal et honnêtement je suis souvent réservé avec ce genre d’annonce mais là autant voir dire que même moi je suis sur le cul hein je vous explique ça rapidement
en gros, ils ont publie des preuves dans l analyse harmonique avec la conjecture de lissage local de sogge et la geometrie des ondes), la physique théorique avec l’existence de theories quantiques des champs en interaction avec jauge et fermions et la theorie de la complexité algorithmique mais la ou le sol a vraiment tremble a mes yeux c’est sur ce dernier point avec la refutation de l’hypothese du temps exponentiel (eth)
pour comprendre le vertige il faut revenir aux fondamentaux et comprd’re que les problemes de classe P sont calculables en temps polynomial par une machine de turing deterministe alors que les problemes NP complets (comme 3-sat) semblaient condamnes a une explosion combinatoire exponentielle où le temps de calcul explose avec la taille des donnees
ce que la preuve etablit via un algo deterministe sur les formules cnf c’est qu’on peut exactement compter les affectations satisfaisables en temps subexponentiel et espace polynomial, ruinant l’idee qu’il existe une barriere mathematique infranchissable et ouvrant un regime algorithmique intermediaire quasi-polynomial  O(n log n) tout a fait inattendu!!!!
si je vous explique tout ça cnest aussi pour vous parler des implications industrielles de cette fissure dans le mur NP dur sont titanesques car en effondrant les tps de calcul sur la recherche operationnelle, on passe du tâtonnement heuristique a la resolution exacte pour le repliement des proteines en biophysique par ex, la verification formelle des puces de silicium sub-nanometriques, le controle magneto-hydrodynamique en tps reel des plasmas de fusion nucleaire dans les tokamaks (iter/spheromak) = ÉNORME CHALLENGE ou l’optimisation combinatoire instantanee des topologies de reseaux quantiques et tant d’autres sujets qui vont vous impactér dans votre vie de tous les jours et qui ont de quoi remettre en question toute votre vision du futur
We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.
openai.com/index/building-sc…
Paul Masurel 🦀 retweeted
blogs
datadoghq.com/blog/ai/harnes…
datadoghq.com/blog/ai/fully-…
sesh-mode
github.com/quickwit-oss/quic…
antithesis skills
github.com/antithesishq/anti…
Paul Masurel 🦀 retweeted
La France en ragebait y’a rien de mieux ptdrrrrrr l’IA c’est tellement méchant 🤣🤣🤣🤣🤣
5 days, reached out through Support, Discord, GitHub and X - and still no @MistralAI answer on what appears to be a serious API bug impacting all models… how come 🤯
Any chance someone in my network or from @MistralAI directly can help with a pretty bad API behavior impacting literally all tested models (github.com/mistralai/client-…).
This was my first experience plugging Mistral on @AltertableAI and I must confess it wasn’t smooth to finally realize this… how come is that not spotted by any kind of tests ?!
Happy to help more, just tell me…
Paul Masurel 🦀 retweeted
10 remarks on AI and math that can be true all at once:
1. The recent progress is stunning and exciting.
2. The Quasi-Riemann Hypothesis is arguably Fields-medal level.
3. Mathematics is being transformed and there is no way back.
4. OpenAI doesn't give a fuck about the future of mathematics.
5. Half-baked unreadable preprints claiming major results can effectively stall mathematical progress.
6. Some mathematicians are responding with naive, overly defensive comments.
7. The social-media wankers framing this as "OpenAI tramples gatekeeping" have no experience of mathematical research.
8. The product of mathematics is clarity and understanding. Not theorems, by themselves.
9. Mathematics is a critical infrastructure of human cognition. The human need for understanding isn't obsolete and it can't be offloaded to machines.
10. We haven't "solved" mathematics and we won't get there before the Sun burns out.
Paul Masurel 🦀 retweeted
🇺🇸
Rare moment de lucidité de la part de Donald Trump :
"C'est incroyable. Je peux dire les choses les plus stupides, et vous applaudissez."
Paul Masurel 🦀 retweeted
Berlin-based search engine Ecosia, used by government services, is dropping its French AI partner, Mistral, its CEO told POLITICO.
The company will now bet on open-source models, including those from China.
vist.ly/5mh5y
Paul Masurel 🦀 retweeted
The math results yesterday are truly incredible.
This one for instance. I spent a ton of time implementing different integer multiplication algos back in school. The time complexity of this new algorithm is so deeply psychotic and unsettling lol.
harper.fast/ made typescript bindings for tantivy.
github.com/HarperFast/fullte…
1
222
Paul Masurel 🦀 retweeted
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.
🤖 Made with AI
Two weeks ago, @PlanetScale launched TIN. They benchmarked it against us, and it was very fast. So we spent those two weeks making ParadeDB faster without their tradeoffs.
Today, we are faster than TIN.
paradedb.com/blog/opening-a-…
wtf
Replying to @dsp_
Hey David, Gayani here from Figma. You're right that our remote MCP server only accepts clients on our supported list, and Pi isn't on it yet. You can see the current list in our MCP catalog at figma.com/mcp-catalog. If you'd like Pi considered for a future addition, please fill out this form: forms.gle/qSvUawwznWyoj8Go7.
1
433
If you have a loooooot of loooogs, and you are interested in Quickwit or what we are building at Datadog, please come!
Hey hey hey, our first BYOC for observability meetup in Paris is coming on Oct 6!
@Dailymotion & @Winamax will share how they run Datadog BYOC Logs, Vector & @Quickwit_Inc at scale.
Talks, then drinks & pizza at @datadoghq Paris 🍕
Register 👇
luma.com/ebotlok7
Paul Masurel 🦀 retweeted
We're running Rust on Laravel Cloud and it's terrifyingly awesome
> Running Rust (Quickwit) on Laravel Cloud enables cheap, scale-to-zero search; polyglot tools now feel natural with strong AI coding help.
Paul Masurel 🦀 retweeted
Hey hey hey, our first BYOC for observability meetup in Paris is coming on Oct 6!
@Dailymotion & @Winamax will share how they run Datadog BYOC Logs, Vector & @Quickwit_Inc at scale.
Talks, then drinks & pizza at @datadoghq Paris 🍕
Register 👇
luma.com/ebotlok7
Paul Masurel 🦀 retweeted
Searching a few million rows is easy. Searching billions in real time without a huge bill is the hard part.
Quickwit (@quickwit_inc) is built for that: fast full-text search over massive datasets, with real-time indexing baked in.
openalternative.co/quickwit
I get the argument, but I look at assembly too.
“Why would I look at code? It’s like assembly, like a compiled artifact.”
Sensei @karpathy fully agentic, 18m after “I don’t really use autocomplete AI code tools”


























