Google a lancé Gemini 2.5 Flash, un modèle de raisonnement qui rivalise avec o4-mini (actuellement en aperçu). Pensé comme un modèle "économe en réflexion", il peut gérer jusqu’à 24 000 tokens tout en trouvant un équilibre entre qualité des réponses, coût et rapidité. Ses
MACHINE LEARNING
-
Profluent Releases ProGen3 Models for Advanced Protein Design
By
–
Biotech co Profluent dropped ProGen3 models for complex protein design
— Rowan Cheung (@rowancheung) 21 avril 2025
—Flagship 46B LLM trained on 3.4B+ protein sequences
—Designed antibodies matching approved therapeutics
— Also created gene editing proteins half the size of CRISPR-Cas9pic.twitter.com/XbEpePIpdRBiotech co Profluent dropped ProGen3 models for complex protein design —Flagship 46B LLM trained on 3.4B+ protein sequences
—Designed antibodies matching approved therapeutics
— Also created gene editing proteins half the size of CRISPR-Cas9 -
Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, Applications
By
–
A Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, and Applications https://
arxiv.org/pdf/2503.07137 -
Antidistillation Sampling Protects Models Against Knowledge Theft
By
–
Antidistillation Sampling — a novel method to curb the effectiveness of distillation without sacrificing model performance.
A technique big AI labs might be very interested in.
Paper: https://
arxiv.org/pdf/2504.13146
Page: https://
antidistillation.com -
ChatGPT o3 and Gemini Deep Search surpass Perplexity
By
–
ChatGPT o3 / Gemini Deep Search > Perplexity
-
LangChain Korean Tutorial: Educational Resources and Interactive Examples
By
–
LangChain Korean Tutorial A Korean-language educational resource delivering LangChain tutorials through ebooks, YouTube content, and interactive examples. Covers core LangChain concepts, LangGraph systems, and RAG implementations. Explore the full tutorial here
-
o3 leads in intelligence, 4o for cost and speed
By
–
o3 definitely is the smartest kid on the block right now! 4o is still good for cost efficient things (when you use API if you're a developer) and faster things like if you need really a small task solved fast. But o3 intelligence is higher.
-
Ultimate GPT Wrapper Blueprint for Wealth Creation
By
–
Steal my prompt to build GPT wrappers that turn complex tools into simple money-makers non-technical people will happily pay for. —————————————–
ULTIMATE GPT WRAPPER BLUEPRINT
—————————————– ## WEALTH CREATION THROUGH AI -
RL, Attention, and Bayesianism Alone Insufficient for AI Progress
By
–
RL is not all you need, nor attention nor Bayesianism nor free energy minimisation, nor an age of first person experience. Such statements are propaganda. You need thousands of people working hard on data pipelines, scaling infrastructure, HPC, apps with feedback to drive
-
Foundations of Large Language Models by Xiao and Zhu
By
–
Foundations of Large Language Models Tong Xiao, Jingbo Zhu: https://
arxiv.org/abs/2501.09223 #ArtificialIntelligence #DeepLearning #MachineLearning