Small Language Models are the Future of Agentic AI! NVIDIA Research just dropped a paper that could change how you think about building agentic systems. Here is everything you need to know:
RESEARCH
-

AI Progress Continues Exponentially Without Unexpected Breakthroughs
By
–
The current state of AI for sustained work: exponential progress continues with no unexpected leaps but also no walls. (Yes, this METR measure is just one of many benchmarks, and like all benchmarks has flaws, but also has the advantage of have neither a ceiling or floor effect)
-
Navigating the New World of AI Representation and Innovation
By
–
Navigating The New World Of AI Representation
#AI #AIio #AIInnovation #ML #DataScience #Futureofwork @HaroldSinnott @fogoros @iainljbrown @NandoDF @katecrawford @drhassanrashidi @YuHelenYu http://
ow.ly/je3a30sOj7C -
Alibaba Launches Wan 2.2 Open Source MoE Video Model
By
–
C’est fou…
— VISION IA (@vision_ia) 9 août 2025
Alibaba vient de lancer Wan 2.2, le tout premier modèle vidéo open source à architecture MoE avec contrôle cinématographique !
Une mise à jour majeure : qualité cinématographique améliorée, mouvements plus fluides et suivi de prompt optimisé.
Exemples : pic.twitter.com/H3GkT8ED9aC’est fou… Alibaba vient de lancer Wan 2.2, le tout premier modèle vidéo open source à architecture MoE avec contrôle cinématographique ! Une mise à jour majeure : qualité cinématographique améliorée, mouvements plus fluides et suivi de prompt optimisé. Exemples :
-
LLMs Are Web Interface Alternatives Not Proto-AGI
By
–
LLMs aren't proto-AGI. They're just an alternative interface to the Web.
-

GPT Suffers Severe Schizophrenia: Critical Analysis
By
–
If GPT is a brain, it suffers from severe schizophrenia.
-
GPT-4o Stagnation Ends with o1 Release Acceleration
By
–
There definitely was a gear shift with RL. I remember we were stuck with 4o for like 10 months with tiny updates about a year ago, until o1.
-
Genie 3 advances showcase AI potential with computational demands
By
–
Y avances como Genie 3, esta misma semana, demuestra que la IA aún tiene muchos trucos que ofrecernos. Eso sí, claro, siempre que le demos computación para comer 🥄https://t.co/vkA8VZ8F4S
— Carlos Santana (@DotCSV) 9 août 2025Y avances como Genie 3, esta misma semana, demuestra que la IA aún tiene muchos trucos que ofrecernos. Eso sí, claro, siempre que le demos computación para comer
-
GPT-5 differs from GPT-4.5 in computational scale and training approach
By
–
GPT-5, a diferencia de GPT-4.5 no es un modelo mucho más grande que su antecesor. La saga GPT siempre se ha caracterizado por entrenar transformers cada vez más grandes y costosos computacionalmente, pero en esta ocasión ha sido diferente. En parte tiene que ver porque el
-

AI Models Approach Advanced Reasoning With Proper Encouragement
By
–
With encouragement to actually think, it gets sooo close!