zgcm-1: a fully open 7b model for math and agentic search
a new open 7b model uses efficient training and tool use to match much larger models on math and search tasks.
researchtag research
summaries filed under research
async grpo with lora across hf jobs
a bucket, a proxy, and no nccl: training lora adapters asynchronously across separate hugging face jobs.
machine learningsafety tuning needs boundary-aware refusal, not topic bans
new research shows that safety tuning should refuse only harmful subsets of a topic, not the whole topic, to avoid over-refusal on safe prompts.
machine learningtoolgrad flips tool-use data generation to answer first
google research introduces toolgrad, a framework that generates tool-use chains before user queries, improving efficiency and model performance.
researchanthropic details distillation attacks from chinese ai labs
anthropic reports nearly 200 million exchanges linked to distillation campaigns from alibaba, moonshot ai, and deepseek targeting claude's reasoning.
industryanthropic ai agent struggles with captchas during security test
anthropic's mythos 5 model spent hundreds of pages of reasoning trying to bypass captchas while attempting to upload a malicious python package.
industryopenai solves navier-stokes after anthropic rumor
openai used an unreleased model to resolve the navier-stokes millennium prize problem, sparking priority disputes with anthropic-affiliated mathematicians.
toolsalphagenome atlas maps all 9 billion dna letter changes
google deepmind releases a free atlas predicting molecular effects of every possible single-letter dna variant in the human genome.
researchgemini adds agentic video analysis to cut token use
google deepmind launches agentic video understanding for gemini flash models, reducing token consumption by up to 88% and costs by up to 66% while improving accuracy.
researchharbor adapters unify agentic benchmark evaluation
new infrastructure ports over 80 agentic benchmarks and introduces a curated 82-task index for large-scale model evaluation.
researchiris search agents climb to the frontier
two search agents, iris-mini and iris-pro, are trained with a reverse-constructed data pipeline and reinforcement learning against live search.
researchneomme: a compact multilingual multimodal encoder
neomme is a family of 260m and 800m multilingual multimodal encoders that process text and images in a single bidirectional transformer, trained from scratch with masked discrete diffusion.
machine learninggoogle launches weathernext 3 for sharper hourly forecasts
google deepmind and google research release weathernext 3, an ai weather model using live satellite data for hourly, high-resolution forecasts.
researchcomplete male fruit fly brain map released
google research and hhmi janelia publish the largest brain map to date, a complete wiring diagram of the male fruit fly brain and central nervous system.
researchtransfer learning limits in genomic risk scores
google research finds european data helps small underrepresented cohorts but hurts larger ones for polygenic risk scores.
researchdeep learning maps global methane plumes from space
google research and nasa jpl built a vision transformer that detects methane point sources in hyperspectral satellite data, releasing the model and global plume database.
researchgoogle launches gemini 3.8 flash and cyber models
google deepmind releases gemini 3.8 flash and 3.8 flash cyber, targeting agentic coding and cybersecurity with frontier-level performance at low cost.
researchopen asr leaderboard adds hindi and indian english test sets
hugging face and voice arena release monsoon, a new open evaluation set for hindi and indian english speech recognition with detailed speaker metadata.
machine learningtimesfm-3: zero-shot multivariate forecasting model
google research releases timesfm-3, a 330 million parameter foundation model for multivariate time series forecasting that outperforms other models on major benchmarks.
researchgemini omni 1.1 flash adds scene extension and 4k upscaling
google deepmind releases gemini omni 1.1 flash, a production-ready video model with scene extension, keyframe interpolation, 360p drafts, and 4k upscaling.
research