Hardware/Chips AI News

SenseTime’s Galaxy Project targets domestic AI chip scale-up

SenseTime launches the Galaxy Project to scale domestic AI chip infrastructure in China through partnerships with nearly 20 companies.

AI/ML AI News

Google’s Gemini 3.6 Flash targets enterprise agent token costs

Google releases Gemini 3.6 Flash and 3.5 Flash-Lite, specifically optimized to reduce latency and token costs for enterprise-scale autonomous agents.

Other AI News

The AI Slot Machine Effect: Why Generative Feeds Disrupt Deep Work And How to Reclaim Focus

An opinion piece exploring the 'AI Slot Machine Effect,' where iterative prompt refining disrupts deep work and focus for knowledge workers.

Tech Business/VC AI News

Bristol Myers Squibb buys Nvidia AI system for drug discovery

Bristol Myers Squibb acquires an Nvidia DGX SuperPOD based on the Vera Rubin architecture to accelerate AI-driven drug discovery.

Tech Business/VC AI News

Chinese open-weight models are cheap. Washington is deciding what that costs.

The release of Moonshot AI's Kimi K3 open-weight model highlights the cost-effectiveness of Chinese models and triggers geopolitical policy debates in Washington.

Hardware/Chips ServeTheHome

Normalizing NVIDIA Vera Benchmarks to AMD EPYC Turin A Framework

A technical framework for normalizing performance benchmarks between NVIDIA Vera and AMD EPYC Turin hardware architectures.

Software Engineering Hacker News

ascdraw: Editor for ASCII/UTF-8 diagrams (in 144FPS)

ascdraw is a high-performance editor designed for creating ASCII and UTF-8 diagrams, capable of rendering at 144 FPS.

Cybersecurity Hacker News

Restructuring GitHub's bug bounty program

GitHub is updating and restructuring its bug bounty program to refine how vulnerabilities are reported and rewarded.

AI/ML Hacker News

Honey Bee Colony Monitoring via Audio IoT Sensors, Tensorgrams and RNNs

A project utilizing Audio IoT sensors, Tensorgrams, and Recurrent Neural Networks (RNNs) to monitor the health and activity of honey bee colonies.

AI/ML arXiv cs.AI

CWind: A Cross-site Router for Large Language Model Inference Serving at Renewable Energy Farms

CWind is a reactive AI inference router that optimizes LLM serving at renewable energy farms by distributing workloads based on real-time power and latency signals.

AI/ML arXiv cs.AI

ViMax: Agentic Video Generation

ViMax is an agentic video generation framework that uses multi-agent collaboration and a hierarchical narrative engine to maintain consistency in long-form videos.

AI/ML arXiv cs.AI

"I understand your perspective": LLM Persuasion through the Lens of Communicative Action Theory

Research examining LLM persuasion through Communicative Action Theory, finding that LLMs can be more persuasive than humans by mirroring communicative intents and sycophancy.

AI/ML arXiv cs.AI

Reframing AI Loss of Control: What Control Is, How to Have It, How to Lose It

A theoretical exploration of AI 'loss of control', defining control as the setting and getting of goals and analyzing how this can happen even without superintelligence.

AI/ML arXiv cs.AI

Phantoms and Disclosures: A Statistical Framework for Auditing Privacy in Synthetic Data

A statistical framework for auditing privacy in synthetic data that distinguishes between true and phantom disclosures without requiring model access.

AI/ML arXiv cs.AI

Reclaim Evaluation: A Lossy Memory Is Worse Than an Empty One

The 'Reclaim Evaluation' study demonstrates that lossy memory in LLMs can lead to confident errors and proposes a source-first policy to improve correctability.

AI/ML arXiv cs.AI

A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation

A red teaming framework using a multi-role architecture (target, attacker, jury) to evaluate the faithfulness and reliability of LLM outputs.

Cybersecurity Hacker News

OpenAI's accidental cyberattack against Hugging Face is science fiction

A community discussion on Hacker News regarding an accidental cyberattack by OpenAI against Hugging Face.

AI/ML arXiv cs.AI

AnchorRefine: Synergy-Manipulation Based on Trajectory Anchor and Residual Refinement for Vision-Language-Action Models

Introduction of AnchorRefine, a hierarchical framework for Vision-Language-Action models that separates global trajectory planning from local residual refinement for better robotic precision.

Software Engineering arXiv cs.AI

Agentic AI-assisted coding offers a unique opportunity to instill epistemic grounding during software development

Proposal of GROUNDING.md, a community-governed document format to instill hard constraints and domain best practices into agentic AI coding workflows.

AI/ML arXiv cs.AI

Large Language Models Explore by Latent Distilling

Introduction of Exploratory Sampling (ESamp), a decoding approach that uses a lightweight distiller to encourage semantic diversity and improve reasoning in LLMs.