SuperGLUE
PulseAugur coverage of SuperGLUE — every cluster mentioning SuperGLUE across labs, papers, and developer communities, ranked by signal.
-
New JIVEAdapter method improves LLM fine-tuning efficiency
Researchers have developed JIVEAdapter, a novel method for parameter-efficient fine-tuning of large language models. Unlike existing single-task adapters, JIVEAdapter decomposes weight updates into a shared 'Joint' stru…
-
New READ method simplifies combining LLM adapter skills
A new research paper introduces READ (Read-only Expansion of Adapter Deltas), a method designed to improve the composition of multiple Low-Rank Adaptation (LoRA) skills for large language models. READ addresses the inte…
-
AI and Infrared Imaging Offer Radiation-Free Pediatric Skeletal Trauma Diagnosis
Researchers have proposed a novel approach combining infrared (IR) imaging with artificial intelligence to create a radiation-free alternative for diagnosing pediatric skeletal trauma. This method utilizes various IR sp…
-
Robots use pretrained vision models for dynamic obstacle avoidance
Researchers have developed a novel method for robots to dynamically avoid obstacles in unstructured outdoor environments without requiring extensive robot-specific training data. The approach utilizes a pretrained visio…
-
General LLMs now outperform specialized clinical AI on benchmarks, but safety concerns persist
General-purpose large language models are now achieving performance levels comparable to or exceeding specialized clinical AI systems on various benchmarks, including those for structured knowledge and reasoning. For in…
-
Computer vision system tracks fish behavior for aquaculture welfare
Researchers have developed a novel computer vision system to monitor fish behavior in aquaculture settings. The system uses object detection and stereo-vision techniques to track individual fish and estimate their 3D po…
-
PEML method optimizes LLM prompts and weights for multi-task learning
Researchers have introduced PEML, a new method for parameter-efficient multi-task learning in large language models. PEML optimizes both continuous prompts and model weights simultaneously, addressing limitations of exi…
-
Google DeepMind releases T5Gemma encoder-decoder LLMs adapted from Gemma
Google DeepMind has introduced T5Gemma, a new family of encoder-decoder large language models derived from their existing Gemma 2 models. This adaptation technique allows for flexible combinations of encoder and decoder…