Sam's News — ai — 2026-09-18¶
AI¶
6.5 Tsinghua Professor's Stealth LLM Startup Reaches $1.4 Billion Valuation¶
A secretive large language model startup founded by a Tsinghua University professor achieved a $1.4 billion valuation in funding.
Sources: The Information RSS
6.5 Scientists assess whether China's latest AI model matches U.S. competitors¶
Researchers evaluate whether China's newest large language model achieves performance parity with leading U.S. AI systems.
Sources: Nature Web Search
6 NVIDIA releases AIPerf benchmark for LLM inference at scale¶
NVIDIA published AIPerf, a new benchmarking tool for evaluating large language model inference performance at scale.
Sources: NVIDIA Developer RSS
6 New AI model from ChatGPT inventor attracts developer interest¶
A new kind of AI model from a ChatGPT inventor is gaining developer attention.
Sources: techcrunch.com Web Search
5.5 PrismML launches compact language model targeting mainstream AI adoption¶
PrismML has developed a small-scale language model with the goal of transforming how consumers and businesses adopt and use artificial intelligence.
Sources: TechCrunch RSS
5.5 Study reveals LLM watermarking can alter AI agent behavior and tool use¶
Research shows that watermarking techniques applied to large language models can inadvertently modify agent behavior and tool-calling patterns.
Sources: analyticsindiamag.com RSS
5.5 ChatGPT co-inventor launches non-verbal AI amid high demand¶
A ChatGPT co-inventor has built an AI system without voice capabilities that is rapidly gaining market adoption.
Sources: autogpt.net RSS
5.5 Best open-source agent harnesses for local LLMs in 2026¶
A roundup of the best open-source agent frameworks and tools for running large language models locally.
Sources: marktechpost.com Web Search
Hardware¶
6.5 GMKtec Unveils EVO-X5 Pro Desktop AI Supercomputer with AMD Ryzen AI Max Processor¶
GMKtec introduced the EVO-X5 Pro at IFA 2026, claiming it as the world's first desktop AI supercomputer capable of running 300B-parameter LLMs fully offline.
Sources: Barchart.com RSS, Business Wire RSS, 01net RSS, TMX Newsfile RSS
AI Safety¶
6 AI Watermarking Could Undermine Reliability of LLM Safety Guardrails¶
Researchers warn that watermarking techniques for large language models may make adherence to safety guardrails unpredictable and difficult to verify.
Sources: Yahoo Tech RSS