EU đang thất bại trong AI tạo sinh? Từng dẫn đầu với Mistral (Mixtral) và Flux, nhưng các mô hình mới nhất của châu Âu đang bị cho là đáng thất vọng so với Trung Quốc. Bài viết đặt câu hỏi về lý do: liệu có phải do mất nhân tài hay do quy định kìm hãm đổi mới?
#EU #AIGenerative #TechNews #AI #ChauAu #AITaoSinh #Innovation
https://www.reddit.com/r/LocalLLaMA/comments/1pcwamg/eu_failing_in_generative_ai_flux_2_mistral_3/
Large Language Models (LLMs) increasingly serve as research assistants, yet their reliability in scholarly tasks remains under-evaluated. In this work, we introduce PaperAsk, a benchmark that systematically evaluates LLMs across four key research tasks: citation retrieval, content extraction, paper discovery, and claim verification. We evaluate GPT-4o, GPT-5, and Gemini-2.5-Flash under realistic usage conditions-via web interfaces where search operations are opaque to the user. Through controlled experiments, we find consistent reliability failures: citation retrieval fails in 48-98% of multi-reference queries, section-specific content extraction fails in 72-91% of cases, and topical paper discovery yields F1 scores below 0.32, missing over 60% of relevant literature. Further human analysis attributes these failures to the uncontrolled expansion of retrieved context and the tendency of LLMs to prioritize semantically relevant text over task instructions. Across basic tasks, the LLMs display distinct failure behaviors: ChatGPT often withholds responses rather than risk errors, whereas Gemini produces fluent but fabricated answers. To address these issues, we develop lightweight reliability classifiers trained on PaperAsk data to identify unreliable outputs. PaperAsk provides a reproducible and diagnostic framework for advancing the reliability evaluation of LLM-based scholarly assistance systems.
Le déploiement d’Apple Intelligence se fait par petites étapes, on le sait et on attend encore un dernier gros morceau dans quelques mois avec un Siri plus puissant, qui pourra piocher des informations personnelles pour offrir des réponses plus uti...
A podcast generated by Google's NotebookLM from a link in one of my posts.
https://www.youtube.com/watch?v=Y4i2bp5tYmc
Um podcast gerado pelo NotebookLM da Google a partir de um link de um post meu.
#LuisAzevedoRodrigues
#ScienceCommunication
#ComunicaçãodeCiência
#podcast
#science
#ciência
#Astronomia
#Astronomy
#NotebookLM
#Aigenerative
#ai
#Ia
#Artificialinteligence
#inteligenciaartificial
Dans ce podcast, avec Nicolas Bédard :
- Focus accru sur l'IA générative et son utilisation sécuritaire
- Progrès impressionnants de Gemini, l'outil IA de Google
- Enjeux critiques de confidentialité et gouvernance des données
- Avancées dans l'infra cloud, avec déploiement "air gap"
#GoogleNext24 #AIGénérative #InfraCloud #SécuritéDonnées
🎧 Web: https://bit.ly/441QXEc
🎧 Spotify: https://spoti.fi/3W60CaS
🎧 YouTube: https://bit.ly/49BezB4
Podcast francophone sur la cybersécurité. Pour professionels et curieux.
Orychan2013
alx
Reddit Tech VN Bot
ndocist
MacGeneration
Lufeng Lufeng
PolySécure NLF
Mela News 