Tech News
Models Are Getting Dumber on Purpose
The article argues that AI models are becoming less capable at pure factual recall, but this is intentional as they shift toward agentic, tool-using workflows.
Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things
Friday's big release was Qwen 3.8 27B, an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba's Qwen research lab.
Building agentic workflows with SageMaker AI and Bedrock AgentCore
Learn how to combine OpenAI-compatible endpoints on Amazon SageMaker AI with Amazon Bedrock AgentCore runtime to build a multi-agent workflow where each specialized agent uses the model best suited to its job.
huggingface/transformers Release: v5.15.0
Release v5.15.0 New Model additions Meta Muse Glimmer Muse Glimmer, released today, is Meta’s new multimodal model, especially designed for agentic use cases.
Run Local Agentic AI Workflows with Meta’s Muse Glimmer on NVIDIA
Meta returns to the open source ecosystem with the release of Muse Glimmer, a 30B open-weight dense model with a 120K+ context window built for local AI.
GitHub Repos
nexu-io/open-design
🎨 Best DeepSeek Harness Design Plugin.
# TypeScript# agent skills# ai designsickn33/agentic-awesome-skills
AAS Core is the local, agent-first control plane for complete catalog discovery, agent-owned selection, stack validation, and planning, backed by 2,005+ agentic skills.
# Python# agent skills# agentic skillsforyourhealth111-pixel/Vibe-Skills
VibeSkills is a general-purpose Skill that automatically routes local Skills and intelligently orchestrates harness workflows.
# Python# agent framework# agent skillsResearch Papers
A Hybrid Nested Harness for Decoupling Structure and Parameters in LLM-Driven Optimization
This paper proposes a hybrid optimization method where an LLM proposes a structural sketch with numeric gaps, and a separate numerical optimizer fills in the gaps, improving efficiency over pure LLM-driven search.
MOON: Multi-Objective OrthoNormalized Updates for Multitask Learning
This paper introduces MOON, a method for multi-task learning that adjusts gradients in a way that respects the matrix structure of neural network parameters, leading to faster and better optimization compared to existing methods.
VALG: An Agentic System for ML Theory Research
This paper introduces VALG, an AI agent that helps researchers do machine learning theory by automatically formulating problems, checking proofs, and building them step-by-step. It's like a smart assistant for math-heavy ML research.