We are working to make them much more token efficient, stay tuned 🙂
LLMS
-
New pricing model for reasoning-enabled AI language models
By
–
As mentioned at the bottom, we chose to have a “thinking off” option with a much lower price specifically so that devs moving from 2.0 had a more clear migration path. But with reasoning is definitely more expensive, very different class of models.
-
Flash 2.0 Model Reaches Stable GA Release Status
By
–
what changed on 2.0 flash? the model is stable and GA
-
Qwen3 32B Delivers Lightning-Fast Performance on SambaNova Cloud
By
–
Speed Alert! @Alibaba_Qwen's Qwen3 32B is blazing fast on SambaNova Cloud! 🔥
— SambaNova (@SambaNovaAI) 8 mai 2025
🗣️ Juggle 100+ languages faster than you can say "Bonjour, 안녕하세요, 你好!"
🧠 Tackle complex agent apps like a caffeinated supercomputer.
Don’t miss out ➡️ https://t.co/zm6RCXXsaPSpeed Alert! @Alibaba_Qwen
's Qwen3 32B is blazing fast on SambaNova Cloud! Juggle 100+ languages faster than you can say "Bonjour, 안녕하세요, 你好!" Tackle complex agent apps like a caffeinated supercomputer. Don’t miss out http://
cloud.sambanova.ai -
Absolute Zero: Reinforced Self-play Reasoning Without Training Data
By
–
Absolute Zero: Reinforced Self-play Reasoning with Zero Data Zhao et al.: https://
arxiv.org/abs/2505.03335 #ArtificialIntelligence #DeepLearning #MachineLearning -
AI Weekly: Google I/O, Mistral Medium 3, Figma, Apple, OpenAI
By
–
Ma dernière newsletter :
– Google prépare des annonces majeures concernant les E/S
– Mistral Medium 3 surpasse ses rivaux
– Figma lance un outil de codage d'ambiance basé sur l'IA
– Apple envisage une recherche par IA pour Safari
– OpenAI s'associe aux gouvernements du monde -
Grok voice mode adds new personality via system prompt update
By
–
BREAKING : @gork will arrive to the Grok voice mode as a new personality! System prompt below
-
Google Developing Feature to Share Custom AI Gems
By
–
Google is finally working on an option to share Gems! It is yet unclear if this feature will be available to personal accounts but it will for sure make Gems much more useful.
-
Online Learning: The Missing Paradigm in Modern AI Systems
By
–
a paradigm of modern AI that feels very behind is that model weights are trained every few months, but frozen between and during interactions real intelligent systems (such as human brains) update their parameters online maybe we'll get there in a couple years
-
Using LLMs for startup idea and competitive moat validation
By
–
4. Find Competitive Moat
— God of Prompt (@godofprompt) 8 mai 2025
Prompt: "Help me validate a startup idea by running it through the 7 Powers framework (Thiel/Helmer): what kind of moat could this idea realistically build?" pic.twitter.com/T68sldr9gj4. Find Competitive Moat Prompt: "Help me validate a startup idea by running it through the 7 Powers framework (Thiel/Helmer): what kind of moat could this idea realistically build?"