Unit 42 identified a Chinese-speaking threat actor using the Hermes Agent framework with DeepSeek as an autonomous offensive operator to target infrastructure via seven vulnerabilities, combining AI-d…
A developer built HOUJIAYAN API, an OpenAI-compatible gateway that routes requests across 16 Chinese LLMs including DeepSeek, Kimi, Qwen, MiniMax, and Doubao. The gateway handles authentication, billi…
Researchers introduced a model-side aggregation interface using Hash-based HyperLogLog (HLL) sketches that maintains compact 2 KiB states alongside frozen language models, achieving 1.6% mean relative…
Perihelion, a 13 MB Rust-based coding agent compatible with Claude Code plugins and skills, launches with support for multiple API providers including DeepSeek, GLM, Qwen, and Anthropic, achieving 98%…
Anthropic, the only major frontier AI lab that refused to sign an Nvidia-led open letter defending open-weight AI models, now argues it opposes a ban on such models while advocating restrictions that …
Google Cloud will retire Moonshot AI's Kimi K2 Thinking model endpoint on October 21st, after deprecating it on July 21st as part of a broader endpoint retirement that includes models from DeepSeek, Z…
Chinese students like 18-year-old Guo Xinyan are increasingly using AI chatbots from Alibaba, ByteDance, Tencent, and Baidu to help select colleges and majors after the gaokao exam, as families seek f…
OpenCode, an open-source coding agent, can now run entirely locally using QVAC, an SDK and CLI from Tether that serves models on-device across GPU backends. The setup eliminates cloud dependencies, gi…
China's AI advantage lies in affordability and accessibility rather than peak performance, with Chinese models processing 18 trillion tokens weekly on OpenRouter by June, triple the U.S. total, accord…
Epilude released Model 4.1, a refined on-device cleanup model for Local Mode on Mac, that reduces critical errors from 4 to 3 and passes 79 of 90 internal benchmark scenarios, up from 78. The model is…
A developer stretched Qwen 35B's context window from 4,096 to 32,768 tokens on a 12GB RTX 4070 by quantizing the KV cache to q8_0 using llama.cpp flags -ctk q8_0 -ctv q8_0, achieving an 8× context inc…
A study of 45 language models found that appending a two-word confirmation tag like 'right?' to a decision question can shift model agreement by up to 64 percentage points, with newer models showing i…
Fermisense, an AI consultancy, reports that a 9B Qwen specialist fine-tuned for $500 on two RTX PRO 6000 GPUs over three and a half days scored 87.3% on a simulated catalog-review benchmark, outperfor…
AI teams risk production incidents by treating model access without proper environment separation. A developer switching a single configuration value can turn an experiment into an outage. The solutio…
Researchers propose a reference feature atlas for auditing language models, trained once on a reference panel and reused for new targets by fitting only a linear decoder. In tests on five 7-9B instruc…
A developer redesigned their portfolio website by migrating the AI chatbot from OpenAI to Hugging Face, switching the database from MongoDB to Neon PostgreSQL, and implementing streaming responses for…
Anthropic CEO Dario Amodei published the company's formal position on open-weights models Sunday, stating that Anthropic has never advocated for a ban on open-weights models and that open-weights mode…
A practical guide to self-hosting Chinese open-weight AI models such as DeepSeek, Qwen, and GLM finds that open weight does not guarantee easy deployment, with hardware costs driven by total parameter…
Chinese AI lab Moonshot AI is releasing its Kimi K3 model weights for free, a move that challenges US AI giants like OpenAI, Google, and Anthropic by offering a high-performance, low-cost open-weight …
Treasury Secretary Scott Bessent warned on July 21 that Chinese companies using model distillation to learn from American AI systems could face sanctions, including Entity List designations, according…