Auto Added by WPeMatico
Training an LLM to call tools reliably requires datasets that pair user queries with correct tool-use chains. Producing that data at scale has been slow and expensive. A team of researchers from Google, the University of Tokyo, RIKEN AIP, and Tohoku University introduce ToolGrad. The research work inverts the usual pipeline: build a verified tool […]
Long-horizon agents have turned LLM serving into an input-heavy workload. Repeated prefills and million-token contexts leave KV caches that strain HBM, SSD capacity, and bandwidth. DeepSeek AI built its newest release around that exact bottleneck. DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with 552B backbone parameters, 196B additional Engram parameters, and a 1M-token context window. It
Google DeepMind has released AlphaGenome Atlas, a catalogue of precomputed predictions for the molecular effects of every possible single-nucleotide variant in the human genome. That is roughly 9 billion single-letter changes. The release also introduces the AlphaGenome Variant Impact (AVI) score, a single number that ranks variants by predicted impact, plus per-variant feature attributions and
OpenBMB has released MiniCPM5-2B, the second checkpoint in the MiniCPM5 series and the follow-up to MiniCPM5-1B. It is a dense causal language model with 2,516,756,480 parameters, of which 1,981,982,720 sit outside the embeddings. It uses 42 layers, grouped-query attention with 16 query heads and 2 key/value heads, and a native context window of 131,072 tokens.
Robot manipulation datasets have grown far slower than the models trained on them, mostly because collection stays closed and centralized. Expert operators gather demonstrations on lab hardware, process them offline, and ship a fixed benchmark that never grows again. A research team from Axis Robotics, UC Berkeley, Georgia Tech, NTU… is proposing a different shape
Most visual document retrievers in production today are hand-me-downs. ColPali and the models that followed it take a generative vision-language model and repurpose it as an encoder. The result still carries a separately pretrained vision tower and a causal decoder that never generates a token. That is parameter and compute overhead for a task that
AI research agents can already propose, implement and score their own machine learning experiments. Idea generation is cheap; verification is not. Training one candidate can consume hours to days of GPU time, so an agent proposes far more candidates than it can afford to run. Which ones get run is the real lever on research
AI weather models have spent three years closing the gap with physics-based forecasting, but two problems stayed open: resolution too coarse for local terrain, and initialization tied to numerical weather prediction (NWP) analysis that arrives about six hours late. WeatherNext 3, released by Google DeepMind and Google Research, attacks both. It takes a live global
Quantitative research agents that write their own experiments can corrupt the evidence they later learn from. A leaky feature that scores well gets stored as a successful precedent and propagated through later iterations. Prompt-level instructions and reviewer agents do not close this, because author and reviewer share the same blind spots. A team of researchers
A team of researchers from Google Cloud AI Research, Washington University in St. Louis and UNC Chapel Hill has released EnvHarness, a programmable layer that turns a static agent benchmark into one that adapts to the policy training on it. LLM agents now learn less from curated text and more from interactive environments, but those
