Fuite : le score de Gemini 3.0 au test Humanity’s Last Exam a été repéré dans le code source. Gemini 3.0 – 32,4 %
Grok 4 sans heavy – 23,9 % Ce benchmark laisse entendre que Gemini 3.0 pourrait être proche de sa sortie… et que GPT-5 pourrait faire face à une vraie
LLMS
-

Leaked Gemini 3.0 Benchmark Score Outperforms Grok 4 on HLE
By
–
-

Developer Reverses GPT-OSS Reinforcement Learning, Releases Base Model
By
–
figured out how to "undo" the RL and turn gpt-oss back into a base model will drop the weights tomorrow gn
-
TorchTitan MoE Implementation with FSDP and Torch Requirements
By
–
torchtitan has an MoE impl that supports grouped mm and composes with FSDP: https://
github.com/pytorch/torcht
itan/blob/main/torchtitan/models/moe.py
… needs the latest torch version though (2.8) which flash-attn doesnt have a wheel for yet 🙁 -
100+ Free Open Source AI Agents and RAG Tutorials
By
–
Stay tuned for more such interesting posts → @Saboo_Shubham_ I have created 100+ AI Agents and RAG tutorials, 100% free and opensource. P.S: Don't forget to star the repo to show your support
-

LangChain Launches Deep Agents: Open Source AI Tool
By
–
LangChain literally reverse-engineered Claude Code and Manus AI to build Deep Agents. It's a Python library to turn any LLM into a deep thinking agent with MCP tools. 100% opensource.
-
Google’s LangExtract Library Extracts Structured Data from Text
By
–
LangExtract = Google's Python library for extracting structured information from unstructured text using LLMs:
-
Distributed Training with torchrun and FSDP Made Simple
By
–
It's pretty easy with torchrun, fsdp, etc.
-
ICML Criticized as Band-Aid Solutions for Large Language Models
By
–
ICML should be renamed ICBL (International Conference on Band-Aids for LLMs).
-

Latent Space Podcast: GPT-5, StarGate, AI Engineers & More
By
–
gdb ep @latentspacepod we asked all your burning questions on gpt5, gptoss, stargate, RL, $300m ai engineers and more! thank you @lindsmccallum for helping us capstone the biggest launch week of the year*! (*until devday )
-
AI Benchmarks Will Continue Beyond Humanity’s Last Exam
By
–
“Humanity’s Last Exam” is probably not going to be the last AI benchmark…