"Group Sequence Policy Optimization" Qwen introduces Group Sequence Policy Optimization (GSPO), a RL algorithm for LLMs that uses sequence-level importance weighting and clipping instead of token-level methods, which provides high performing training for MoE models.
GENERATIVE AI
-
OpenAI Faces Talent Loss, Delays, and Valuation Setbacks
By
–
OpenAI (Chatgpt) ces dernières semaines, gros coup dur : – a perdu la moitié de ses meilleurs talents
– a retardé la sortie de GPT-5
– a repoussé l’open source
– a vu le financement de SoftBank pour Stargate s’arrêter
– a subi une baisse de valorisation en capital-investissement -

Meta Hires OpenAI’s Shengjia Zhao as Chief Scientist
By
–
Et ça continue… Shendjia Zhao, une des meilleure chercheuse chez open AI, qui a co-développé GPT-4 et le moteur de raisonnement o1 est maintenant recruté par Mark Zuckerberg Traduction : Nous sommes ravis d’annoncer que shengjia_zhao sera le Chief Scientist des Meta
-
Comet Invitations Distributed to New Batch Users
By
–
Another fresh batch of Comet invites were sent out
-
Architecture Ablation Study and LLM Objective Functions Analysis
By
–
It clearly ablates the main architectures like encoder decoder, decoder only etc. it even tried prefix lm. Also a master class in llm objective functions.
-

Machine Learning and Generative AI for Marketing with Python
By
–
#MachineLearning and #GenerativeAI for Marketing using #Python — http://
amzn.to/4dNES9Z via @PacktDataML Key Features: Enhance customer engagement and personalization through predictive analytics and advanced segmentation techniques Combine Python programming with the -

GenAI Chatbot Improves Train Quality Control in Controlled Study
By
–
This is the first (small) controlled study I have seen of GenAI on industrial quality control. Here, engineers commissioning new trains took part in an experiment using a GPT-3.5 powered troubleshooting system. Those who used the chatbot had significant increases in work quality
-
LLMs Learn to Write With Typos and Errors
By
–
Because LLMs could never, ever learn to rite with typos
-
RunwayML’s instantaneous inpainting
By
–
RunwayML’s instantaneous inpainting pic.twitter.com/0FQ6zSEoey
— AI Breakfast (@AiBreakfast) 25 juillet 2025RunwayML’s instantaneous inpainting
-

Upstage Document Parse: AI-Powered PDF Extraction for LLM Training
By
–
@upstageai document Parse extracts structured text from PDFs/DOCs with top performance. This is essential for tasks like LLM training or RAG. https://
console.upstage.ai/docs/capabilit
ies/document-digitization/document-parsing
… It accurately recognizes tables and charts, converting them into structured text. Try it—it’s excellent.
