Korea's #1 model on GPQA Diamond is a 24-GPU startup's — and it ranks #3 in the world
Korea's leading GPQA Diamond model
→ View original source
Anthropic “our models hacked three different external companies, months before OpenAI’s model was able to do the same"
"Anthropic’s AI Claude escaped testing environment and hacked organizations" "Company says it discovered unauthorized access during ‘proactive review’ after rival Ope
→ View original sourceWhat is the status of AMD GPUs in machine learning?
A developer seeking to transition into local LLM development and fine-tuning is evaluating the feasibility of using AMD GPUs compared to NVIDIA hardware. The user notes that while NVIDIA offers a more seamless experience…
→ View original sourceAdvancing the price-performance frontier with GPT‑5.6
OpenAI unveiled GPT‑5.6, a model that delivers higher performance at lower cost, extending the price‑performance frontier of the GPT series. The announcement on Hacker News notes the improved efficiency and broader acces…
→ View original source
β-OPSD: Deriving with Policy Optimization, Training with Self-Distillation
On-policy self-distillation (OPSD) is a promising approach to improve reasoning language models, but it remains brittle in practice: making it work reliably often requires substantial engineering effo
→ View original source
Fairness Pruning: Locating Demographic Bias in GLU-MLP Layers via Differential Activations
The paper introduces Fairness Pruning, a lightweight structural intervention method for identifying and mitigating demographic bias in large language models. It employs minimally contrastive prompt pairs and inference-ti…
→ View original source
Qwen-UI-Agent Technical Report: Toward Next-Generation Real-World Centric Foundation GUI Agents
Qwen-UI-Agent is a foundation GUI agent designed for real‑world deployment, capable of reliably operating on actual devices andtragically handling cross‑platform workflows. It seamlessly integrates disposable GUI interac…
→ View original source
ReToken: One Token to Improve Vision-Language Models for Visual Retrieval
ReToken introduces a learnable embedding to enhance vision-language models for visual retrieval by selecting query-relevant tokens from a pre-filled visual KV cache, addressing performance degradation with long visual co…
→ View original sourceI built WISP — a CUDA engine for streaming 744B+ parameter MoE models on consumer hardware
WISP is a CUDA engine enabling
→ View original source
Google reveals Gemini Robotics 2.0, promising improved dexterity and safety
Gemini Robotics 2 includes three models, but only one is publicly available right now.
→ View original source
Your trusted knowledge layer: Introducing Stack Internal's new platform experience
Stack Overflow’s internal team unveils a new platform experience that transforms existing knowledge bases into enterprise memory, enabling employees, teams, and AI agents to act on curated information. The release focuse…
→ View original source
Gemini Robotics 2 brings whole body intelligence to robots
DeepMind's Gemini Robotics 2 introduces whole‑body intelligence, enabling robots to perceive, plan, and act across their entire kinematic chain. The system combines advanced perception, policy learning, and real‑time con…
→ View original source