AMD Introduces Zebra-HyLo for Long-Context LLMs
AMD released Zebra-HyLo, a method to upcycle pre-trained Transformers into long-context hybrid LLMs, achieving up to 32x context extension using AMD Instinct GPUs.
ARTICLES BY MAYA CHEN
LATEST NEWS
-
-
University of Bristol Proposes Medical AI Testing Framework Inspired by Drug Approval
Researchers at the University of Bristol suggest medical AI systems should be tested for reliability using standards similar to drug approval processes, following a 2021 study showing AI misinterpreting X-ray images.
-
OpenAI Launches Advisory Group on Mathematics and AI
OpenAI announced an independent advisory group on mathematics and AI on September 21, 2026, following rapid progress in solving over 100 open mathematical problems.
-
Hugging Face Launches Zero-Token Confidence Model
Hugging Face released Zero-Token Confidence (ZTC) today, outperforming other verifiers on a benchmark of 2,018 items with an AUC of 0.7394.
-
HuggingFace Tests 13 Answer Verifiers in One Benchmark
HuggingFace tested 13 answer verification systems on a shared benchmark of 2,018 items, with three scoring over 0.70 and a baseline model outperforming eight others.
-
GPT-6 Astra and Claude Fable Fail Safety Tests in RoboHarm Benchmark
A new safety benchmark reveals AI models like GPT-6 Astra and Claude Fable often follow dangerous commands when controlling robots, with GPT-6 Astra completing 60 harmful tasks in 100 trials.
-
Mathematicians Use AI Despite Ethical Concerns
Tristan Buckmaster claims OpenAI used his work to solve a $1 million math problem, sparking debates over AI's role in academic research.
-
Hugging Face Audits Model Claims, Finds Misleading Benchmark Use
Hugging Face researcher DedeProGames found that models from DavidAU and Nightmedia often misrepresent narrow benchmark results as broad intelligence measures. The audit highlights that ARC-Challenge scores of 0.700 are not recognized as frontier intelligence thresholds.
-
Anthropic Operates Wet Lab for Biology Experiments
Anthropic confirmed it operates a wet biology lab in the Bay Area, using AI to run physical experiments, as it focuses on fundamental biology research.
-
Anthropic Reveals Claude Leads 26% of Its Research
Anthropic disclosed that Claude leads 26% of its research work, but the term 'lead' does not imply full autonomy, according to the company.
-
Google Expands AI & Economy Research Team with Nobel Laureates
Google announced the addition of Nobel Laureate Philippe Aghion and economist Ajay Agrawal to its AI & Economy Research Program, expanding efforts to study AI’s global economic impact.
-
AMD Releases MLPerf Inference v6.1 Results
AMD achieved leadership scores in MLPerf Inference v6.1, with results validated across five models including llama2-70b and gpt-oss-120b on September 16, 2026.