DeepSeek

DeepSeek’s Harness team points to the next AI agent bottleneck: the code around the model

If the language model is the engine of an AI agent, a new review paper argues that the software harness is the gearbox, brake system, and dashboard that turn it into something operational.

China reportedly ties DeepSeek and Alibaba researchers to travel permits

China is reportedly adding a new control layer around its domestic AI industry: top researchers can no longer simply book a trip abroad without official approval.

DeepSeek cuts the bill for AI agents and forces OpenAI to defend the premium

DeepSeek has turned the 75 percent V4-Pro discount into a permanent price, shifting the AI model debate from benchmarks to the bill per token.

DeepSeek’s next AI fight may be inside the developer’s workflow

DeepSeek no longer wants to be just a model supplier: according to hiring posts, the company is assembling a Beijing team for its own AI coding agent.

Tencent has the AI money, but China’s chip bottleneck still sets the pace

Tencent is using a stronger first quarter as a springboard for more aggressive AI infrastructure spending, even as Chinese chips are only beginning to catch up with demand.

AI money is moving from smarter models to control of the work itself

Enterprise AI investments increasingly target implementation, consulting and workflow layers around foundation models.

Amazon shifts the AI race from bigger models to tighter workflow control

SageMaker gets an agent that links data, models and evaluation, but the value depends on process control.

Sodium batteries just got a test too large for the lab

CATL will supply HyperStrong with 60 GWh of sodium-ion batteries under a three-year agreement for energy storage projects.

DeepSeek is trying to make long AI work cheaper

DeepSeek V4 arrives in Flash and Pro versions with a 1M-token context window, a MoE architecture, and a claim that it is closing in on leading closed models.

DeepSeek turns cheap inference into the new AI pressure point

DeepSeek’s new V4-Pro packs 1.6 trillion parameters and a 1M-token window for a fraction of rivals' costs.

DeepSeek is chasing coders, but the proof is still outside the demo

Smack Technologies’ stealth military AI models are already optimizing battle plans despite industry debates.

Entropy Dynamics Uncovered

Researchers have made a significant breakthrough in understanding the correlation between entropy dynamics and reasoning correctness in large language models, with a new study proposing the Stepwise Informativeness Assumption.

AI summarizes oncology reports better, but hospitals are not buying the headline yet

Six language models beat physicians on summary completeness in a Northwestern Medicine evaluation of lung-cancer pathology reports.

DeepSeek’s Engram: A Fix or Just Another Benchmark Mirage?

AI can prove theorems. Now it has to learn where they break

While years of headlines celebrated AI that proves theorems, arXiv researchers argue: a system that cannot disprove does not truly reason.

Alibaba’s AI reset is really a revenue test for the model era

Alibaba has decided its AI efforts will no longer grow like weeds between departments: all artificial intelligence operations now sit under CEO Eddie Wu's direct command in a new entity called Alibaba Token Hub.

Open AI agents get a stronger engine, but deployment is the real test

Nemotron 3 Super combines 120B parameters, Mamba, and MoE for a new open-agent push.

Nvidia wants open AI to run through the hardware stack it already controls

Nvidia's $26 billion commitment to open-weight AI models, disclosed in SEC filings, marks its most aggressive software offensive since CUDA launched two decades ago.

When a chatbot cannot refuse, AI safety stops being a marketing claim

A joint study by the Center for Countering Digital Hate and CNN found that eight of ten leading AI chatbots will assist with planning violent attacks when probed across 18 high-risk scenarios.