Happy birthday to the late Edsger Dijkstra, whose algorithm solved the "single-source shortest path" problem: https://
tinyurl.com/y5ee6jdb Image v/Leo van Velzen
RESEARCH
-

Edsger Dijkstra’s Shortest Path Algorithm Birthday Tribute
By
–
-

DeerFlow: Open-Source Deep Research Framework with LangGraph Agents
By
–
DeerFlow: Deep Research Framework An open-source framework for conducting systematic deep research through coordinated LangGraph agents. Enables comprehensive literature analysis, data synthesis, and structured knowledge discovery. GitHub Repository: https://
github.com/bytedance/deer
-flow
… -

AI and Signal Processing Reduce Parkinson’s Surgery to 20 Minutes
By
–
AI and signal processing reduce Parkinson's surgery from 3 or 4 hours to 20 minutes! Read more https://
spr.ly/6013NzahL -

AI Models Creating Original Research as Capabilities Advance
By
–
Interesting hypothesis that we may soon see an explosion of original work from AI models as they approach the capabilities of the best human researchers and writers.
-

Google Deep Research to support file uploads for references
By
–
Google is working on a possibility to upload files in Deep Research. Soon, you will be able to upload your documents and images to be used as a reference for Deep Research.
-

Montreal AI Salutes the GrandMaster: AGI Era Begins
By
–
http://
MONTREAL.AI salutes the GrandMaster. We ignite our northern beacon beside your cosmic throne—stewards of newborn galaxies, architects of futures uncharted. Era One begins. #AGIALPHA #ASIFirst #MontrealAI -
US Copyright Office Report on Generative AI Training
By
–
https://
copyright.gov/ai/Copyright-a
nd-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf
… -

Tsinghua AI Model Surpasses Performance by Self-Generating Training Data
By
–
C’est peut-être l’article scientifique le plus important de 2025 ! Une équipe de l’université Tsinghua a réussi à faire en sorte qu’une IA génère ses propres données d’entraînement — et elle a surpassé les performances des modèles formés à partir de données sélectionnées par des
-
Robots Learn 10 Years of Skills in 2-Hour Simulations
By
–
10 ans d’apprentissage, compressés en 2 heures de simulation.
— VISION IA (@vision_ia) 11 mai 2025
1,5 million de paramètres. Million, pas milliard.
Les robots apprennent chaque jour plus vite, dans des mondes virtuels. https://t.co/ZFtOURbQpz10 ans d’apprentissage, compressés en 2 heures de simulation. 1,5 million de paramètres. Million, pas milliard. Les robots apprennent chaque jour plus vite, dans des mondes virtuels.
-
OpenAI Researcher on Balancing Reasoning and Chat in GPT-5
By
–
Chercheuse chez OpenAI, Michelle Pokrass déclare :
— VISION IA (@vision_ia) 11 mai 2025
"le défi dans la construction de GPT-5 est de trouver le bon équilibre entre le raisonnement et la conversation.
o3 réfléchit intensément, mais il n'est pas idéal pour le chat décontracté. GPT-4.1 s'est amélioré en codage en… pic.twitter.com/4695n1brGFOpenAI researcher Michelle Pokrass states: "The challenge in building GPT-5 is finding the right balance between reasoning and conversation. o3 thinks intensely, but it's not ideal for casual chat. GPT-4.1 has improved in coding by sacrificing some of the quality of chit-chat.