@pbeyssac @freeMind106 @gchampeau he's right though, why use the X-best (with X > 3) except because they are european.
English
N
3.6K posts
@freeMind106 @gchampeau I believe you simply have a US-biased way of seeing things :)
English
I this "Mistral in only 74th, what a shame" buzz since yesterday with a lot of broken/incomplete analyses with confirmation biases. Enough.
1. The ranking is here. The scores are very close to each other (top model: 1504, Mistral: 1415) arena.ai/leaderboard/te…
Martin Varsavsky@martinvars
Europe’s best AI model, Mistral Large, ranks 74th in the world. Why does Europe do so poorly? Three words: capital, energy, permitting. US labs spend 10x more on compute, pay half the electricity price, and build datacenters in months, not years. Europe produces great researchers then watches them leave for Silicon Valley. Absurd Regulatory and energy policy have consequences.
English
Faut aimer passer devant les tentes de SDF. Perso, je trouve ça indécent
Le Parisien | Paris@LeParisien_75
Un joggeur toutes les trois secondes sous le Pont-Neuf : à Paris, le succès fou de la course à pied à l’heure du déjeuner ➡️ l.leparisien.fr/CfPa
Français
N retweetledi
Houellebecq himself could not have written a more Houellebecqian sentence
KK@kenki_kids
手コキ店で20代前半のオキニとイチャイチャしてシコシコしてもらった後は行きつけの寿司屋へ。最高です。
English
@Sirupsen p99ers should only boast about p99 numbers
oh wait, or is it you showing grace to competitor ? 😉
English
N retweetledi
« Vous n’avez pas déclaré vos impôts depuis 4 ans ?! »
« Oui, j’étais otage français en Iran. »
« Et alors ? C’est une raison ? »
3615 - Hixe@AllianceDroite
La technocratie française😂😂😂
Français
@YouJiacheng There is more discussion below that comment. It is all a bit complicated and a bit evidence back and forth. But there seems to be pretty big evidence that QJL does not help. It certainly does not help in my own benchmarks
English
I was going crazy because I could not replicate TurboQuant. Turns out the community also had issues. The community quickly made adjustments to "make it work", but what they did not realize is that they reimplemented (most of) HIGGS in the process (full HIGGS would be even better)
English
@Tim_Dettmers the only hope is that the whole field gains from the ripple effect of people looking at TurboQuant, but it sucks for so many teams not at big corp labs T_T
English
We in the quantization community could quickly see this and were flabbergastered by the response to TurboQuant.
Whenever I saw TurboQuant on my timeline, I found it hurtful, because the work of other academics who worked so hard was discounted.
English
N retweetledi
N retweetledi
Our entire LLM inference engine with all the GPU kernels and stuff is around ~70K lines of code
What the fuck?
Garry Tan@garrytan
Absolutely insane week for agentic engineering 37K LOC per day across 5 projects Still speeding up
English
CI feels more interesting today than it ever was. Writing code has gotten a lot faster, but this shifts the bottleneck elsewhere. I’m excited about sandboxes as a primitive for massive parallelization of tests.
English
video from CMU is live. should make it clear we're going for it
truly an honour to present in this forum. been watching CMU videos for years and years ❤️
youtube.com/watch?v=pqoRNw…
YouTube
English
N retweetledi
In the United States car seat laws mandating use until age 8 save 57 lives each year at the cost of about 8,000 births.
CJ@UnderSneege
Isofix killed the 3 child family
English
N retweetledi
N retweetledi
@Dorialexander @PernotLeplay well they are moving to every layer then no? doing consulting application, "research" models worse than chinese open source but at least european ...
English
Europe has everything it needs to lead in AI except the last part:
Today it’s easier for our startups to go to the US to scale than to grow from Europe.
But we’re improving: it used to sometimes be cheaper to create a subsidiary in the US than in another EU state. EU-Inc will change that
Dealroom.co@dealroomco
💥 NEW REPORT: State of AI in Europe: The Invisible Giant 🇪🇺 Europe has as much AI talent as the US 📈 AI investment in Europe is at an all-time high 🤖 Europe leads the world in AI adoption 💡 But the biggest paradox: Europe builds the startups that too often get scaled and owned by the US. The gap isn’t talent or entrepreneurship, but growth capital and speed. New report with @Prosusgroup out now.
English
@Dorialexander @PernotLeplay Do you feel that Mistral moved to application layer? i do
English
@PernotLeplay No actual problem goes deeper: entire funding ecosystem has internalized EU is just good enough for the application layer. Exactly the same issue China faced and finally overcame.
(From DeepSeek founder, Liang Wenfeng)
English
N retweetledi
In hindsight all the sandbox stuff is so obvious because containers without AI are basically just contners which makes no sense.
English
@Dorialexander French engineering quality/marketing ratio highest in the world
English
I like how my TL is simultaneously Cursor spending a ton of RL compute to improve on Kimi 2.5 by 0.7% while Lighton saturates a hard search bench on a shoestring.
Antoine Chaffin@antoine_chaffin
We've almost saturated BrowseComp-Plus with a 150M model... ... but this was an old model and I had a lot of ideas to improve the results 🙁 So maybe it's time to kick off a new challenge and see what's the cheapest setup we can solve BrowseComp-Plus with?
English
N retweetledi
Twitter est davantage un lieu de discussion politique que de révélation de vie personnelle. Je me disais que je pouvais partager un peu. Ma fille est née. Je n'ai jamais vécu un bonheur si grand dans un lieu aussi vide que cette maternité. Je pense aux vidéos récentes de Blast et d'autres influenceurs qui répètent que l'encouragement à être parent est une forme d'oppression. Aux jeunes qui les écoutent : ce sont des messagers du malheur, des philosophes de l'égoïsme. Rien d'autre n'importe que l'avenir. Aujourd'hui sert à demain. Il y a des formes d'amour qu'on ne peut connaître si on pense trop à soi. Pensée affectueuse à tous les parents, il fallait me le dire plus tôt que c'est si beau. N'ayons pas honte de le dire à nos adolescents et nos jeunes adultes.
Français
@antoine_moyroud @cortisquared tu as fait un très bon investissement avec Mistral, mais leur business diverge beaucoup des top labs sur les perfs car etre europeen est un unfair advantage
c'est pas en etant SotA OCR (et encore ...) que tu justifies 5bn de valo
Français
La compliance comme business model, c’est un aveux de non-compétitivité.
Mistral AI for Developers@MistralDevs
🚀 Mistral OCR3 (2512) GA on Microsoft Foundry! 🇪🇺 Data Zone support – GDPR/data residency ready 🎯 99%+ accuracy on tables, handwriting, multi-language Process invoices/contracts fast & compliant. #AzureAI #DocumentAI
Français