"Mixture-of-Depths Attention" This paper teaches a Transformer to attend not just across tokens, but also to depth KV from its earlier layers. That helps recover shallow-layer signals that standard residual stacking tends to dilute, improving performance with only a small extra
LLMS
-

Maximum Likelihood RL Framework for Math Reasoning Models
By
–
The current RL setup might not be the right objective for reasoning models In our latest AI4Science talk, Fahim (
@FahimTajwar10
), PhD student at CMU, presented “Maximum Likelihood Reinforcement Learning”, a new framework for binary-reward tasks like math reasoning, navigation, -

Claude OAuth Service Outage Confirmed by Users
By
–
Okay it REALLY isn't just me, it's Claude's OAuth being down So annoying!
-
Managing AI Agent Threads Requires Intense Cognitive Work
By
–
There's worry that people will stop using their brains with LLMs, but managing several AI agent threads in parallel has been some of the most cognitively intensive work I've done in years
-
NVIDIA Releases Open Models for Agentic AI and Robotics
By
–
NVIDIA open models provide the foundation to innovate across agentic AI, robotics, autonomous vehicles, and research. 🤖
— NVIDIA (@nvidia) 18 mars 2026
This release contributes 10 trillion language tokens, 500,000 robotics trajectories, and 100 terabytes of vehicle sensor data to the community. These… pic.twitter.com/ALZ7HKHtpTNVIDIA open models provide the foundation to innovate across agentic AI, robotics, autonomous vehicles, and research. This release contributes 10 trillion language tokens, 500,000 robotics trajectories, and 100 terabytes of vehicle sensor data to the community. These
-

Google Testing Google Play Books Integration for NotebookLM
By
–
Google is working on a possibility to use Google Play Books as a source for NotebookLM. PlaybookLM
-
Gemini ChatGPT Claude Positioned for Distinct AI Market Roles
By
–
Gemini will be the Android / iOS model ChatGPT will be the enterprise model Claude will be the specialized agents model
-

Mistral Small 4 Now Available on Poe Platform
By
–
Mistral Small 4 is live on Poe! One model for reasoning, vision, and agentic coding. Good for things like solving multi-step math or research problems, extracting structured data from images or scanned documents, debugging and navigating a codebase, analyzing a chart and
-
GPT-5.3-Codex-Spark: Three Real Workflows for Building
By
–
what can you build with gpt-5.3-codex-spark?@jxnlco from @OpenAI demos 3 real workflows — ones you can set up yourself inside the Codex app to help you spend less time on overhead and more time building.
— Cerebras (@cerebras) 18 mars 2026
00:09 – what is gpt-5.3-codex-spark?
00:25 – workflow 1: multi-agent… pic.twitter.com/ckpJREPJ3ewhat can you build with gpt-5.3-codex-spark? @jxnlco from @OpenAI demos 3 real workflows — ones you can set up yourself inside the Codex app to help you spend less time on overhead and more time building. 00:09 – what is gpt-5.3-codex-spark? 00:25 – workflow 1: multi-agent daily briefing from slack, drive & meets 01:06 – workflow 2: automated PR review 01:31 – workflow 3: real-time interactive coding 02:56 – what speed changes, and what's coming next
-

Microsoft Copilot to integrate deeper OneDrive context for improved responses
By
–
Copilot may also get a deeper OneDrive sync for Copilot Library. "Copilot will use context from your Microsoft OneDrive to improve answers."
