Data/AI engineer - Everything is compute I repair old fighter jet ✈ too I write pyspark for money
Religiously computing
Joined November 2023
- Tweets9.1K
- Following1.7K
- Followers977
- Likes46.6K
Ce que les gens loue chez mistral ce n'est pas la performance du modèle, c'est surtout le contrôle sur la stack d'inférence.
En Chine, j'étais persuadé que tout le monde utilisait des IA chinoises.
Ou à la limite de l'open source.
En vrai pas du tout.
Ils utilisent ChatGPT et Grok bot à fond. Toute la journée.
Du coup j'ai demandé pourquoi ils passaient par des solutions américaines.
Réponse simple : c'est une question de compétitivité.
Pour eux les outils US sont juste bien au dessus. Donc pas de nationalisme, pas de débat sur la confidentialité.
T'utilises l'outil le plus puissant que t'as à ta disposition. Point barre.
Et à mon avis c'est exactement ce qui va se passer en Europe.
Mistral c'est bien. C'est une belle initiative.
Mais personne ne se souvient du deuxième.
Faut être premier.
You should try to play with ML4 too, its really good
mistral’s student plan is insane right now
$6 for $165 worth of DeepSeek and GLM usage
you get:
> mistral pro at $5.99/mo (normally $14.99)
> $150 vibe code usage + $15 studio api credits
> access to deepseek v4.1 flash, glm 5.3, glm 5.3 flash and more
> full vibe coding (cli, ide, web)
who it’s for:
verified students with an institutional email
how to claim:
1. go to mistral.ai/pricing/ (or chat.mistral.ai/upgrade/stud…)
2. click “i’m a student”
3. sign up / verify with your school email
4. subscribe ($5.99/mo)
5. grab your vibe key and studio api key and start using
only for verified students, up to 12 months. grab it now!!!!!!
Rejoice !
We're going to have a Qwen 4 27b competitor
Z.ai chief scientist Jie Tang just revealed the training pipeline behind the upcoming GLM-5.5 (32B, 122B, and 256B) during a lecture at Tsinghua University today.
Instead of relying on public datasets, their pretraining pipeline pulls from six separate internal engines:
- Web: 1 trillion pages crawled at ~10B pages/day with PB-scale template clustering
- Documents: 2 billion scientific PDFs parsed through GLM-OCR (4x faster with speculative decoding)
- Books: 50 million books with layout reordering and garbled-text filtering
- Code: 500 million repositories turned into executable tasks with real tests and issue descriptions
- Video: 10 billion videos for long-sequence world knowledge
- Audio: 300 million hours rewritten via ASR into trillions of tokens
Having to go back to use Gemini 3.8 flash at work feel like i'm speaking to a monkey 😭
Please Google make argon available in Europe 😭
Overall i'm really happy with the new Hermes mobile app ! I have a few visual glitches but the app is solid
You should definitly give it a go
At some point OpenAI is going to build the ram maximizer
I guess the paperclip maximizer will have to wait
It would be insane if we could do homomorphic inference
Even the most private and sensitive customers would be able to go where the token is the cheapest. And also reduce Capex at the same time
I tried to play with databricks, its really good
coming from foundry its really differents, but at least its a lot cheaper and easier to deploy on privates instances
As we are getting into winter, the main factor limiting my speed on my wheel is no longer my fear, but how much i can chill my balls before it hurt
The wind is getting super cold at high speed
I love it !
just i'm too dumb and tool it in XL instead of L 😭
Thank a lot for the merch @NousResearch
🤖 Made with AI
First time in the business class from KLM on a Embraer 195-E2 (First time in this plane too).
For now its pretty good
>lots of leg room for such a small aircraft
>crew is super nice
>lots less noise
We have internet on board, i can talk to my agents !
and Prime intellect just made me a nice ui to review all the flight i've done (still need to add the older ones)
They use Viasat as a provider and tbh its not bad at all ! obviously the ping is atrocious since its geostationary satellites but i'm impressed
