Semantic Scholar API
PulseAugur coverage of Semantic Scholar API — every cluster mentioning Semantic Scholar API across labs, papers, and developer communities, ranked by signal.
11 day(s) with sentiment data
Semantic Scholar API to be integrated into FrontierCS 2 benchmark tooling
The FrontierCS 2 benchmark aims to evaluate LLM agents on complex R&D tasks. As Semantic Scholar is a primary source for research papers, it's plausible that tools developed for this benchmark will integrate with the Semantic Scholar API to fetch relevant papers and information for the LLM agents to process and build upon.
VLA models' coordinate-binding issues may drive demand for spatial reasoning APIs
Recent evidence shows VLA models struggle with spatial reasoning due to motor programs being tied to absolute coordinates and ignoring scene changes. This brittleness suggests a need for external tools or APIs that can provide more robust spatial understanding, potentially increasing usage of services like the Semantic Scholar API if it offers relevant research or datasets.
Semantic Scholar API usage to increase for research on VLM inverse scaling
The discovery of inverse scaling in VLMs on specific subtasks like multi-image spatial reasoning indicates a complex research area. Researchers investigating this phenomenon may leverage the Semantic Scholar API to find papers detailing VLM architectures, training methodologies, and benchmark results, driving increased API usage.
Semantic Scholar API may see increased usage for VLA and spatial reasoning research
With new benchmarks like FrontierCS 2 and research into VLA model limitations (e.g., absolute coordinate binding) and inverse scaling in VLMs, there's a growing body of complex, multimodal AI research. The Semantic Scholar API is likely to be a key tool for researchers navigating this expanding landscape, potentially leading to increased API calls and data requests related to these specific subfields.
Semantic Scholar API likely used to access SpatialBrain and HOLI-WORLD research
The recent papers on SpatialBrain and HOLI-WORLD, both focusing on enhancing AI spatial reasoning and egocentric learning, were likely accessed and potentially cited via the Semantic Scholar API. Given Semantic Scholar's role as a major academic search engine for AI research, its API is a probable tool for researchers to discover and integrate findings from these cutting-edge projects.
-
Visual foundation models' spatial perception capabilities analyzed in new research
A new research paper explores the spatial state representations within visual foundation models, evaluating their capabilities across four levels: spatial structure, object persistence, predictive spatial updates, and a…
-
CodeSpecWeave framework enhances AI coding agents with interdependent specifications
Researchers have developed CodeSpecWeave, a framework designed to synthesize complex software development tasks with interdependent specifications and verified tests. This approach aims to train coding agents more effec…
-
CEPO framework enhances multi-agent prompt optimization
Researchers have introduced Coordinator-Executor Prompt Optimization (CEPO), a novel framework designed to enhance prompt optimization within multi-agent systems. CEPO distinguishes between shared coordination and agent…
-
New SARL framework boosts LLM search efficiency by recognizing knowledge gaps
Researchers have developed a new framework called Self-Assessment Reinforcement Learning (SARL) to enhance the search capabilities of large language models. SARL enables these models to identify missing information and …
-
SpatialBrain framework learns from experience to boost AI spatial reasoning
Researchers have introduced SpatialBrain, a novel framework designed to enhance spatial intelligence in AI systems by learning from experience. This system separates scene evidence from reasoning processes, utilizing a …
-
FrontierCS 2 benchmark tests LLM agents on computer science R&D tasks
A new benchmark called FrontierCS 2 has been introduced to evaluate the capabilities of LLM agents in contributing to frontier computer science research. This benchmark features 48 open-ended tasks across 13 domains, wh…
-
Study reveals VLA models' motor programs tied to absolute coordinates, ignoring scene changes
Researchers have conducted a mechanistic interpretability study on vision-language-action (VLA) models, revealing that motor programs are often tied to absolute coordinates and disregard changes in the visual scene. Thi…
-
LLM Agents Lose Accuracy When Retrieving Long Tool Outputs, Study Finds
A new paper from Hugging Face investigates how Large Language Model (LLM) agents handle long outputs from scientific tools. When agents retrieve these outputs from files instead of keeping them in context, GPT models li…
-
VideoMaMa++ enhances temporal consistency in video matting
Researchers have developed VideoMaMa++, a diffusion-based framework for video matting that enhances temporal consistency and detail accuracy, particularly for long videos. The system utilizes separate token streams and …
-
Vision-Language Models Show Inverse Scaling on Specific Subtasks
A new research paper reveals that vision-language models (VLMs) can exhibit inverse scaling, meaning larger models sometimes perform worse than smaller ones on specific subtasks. This phenomenon was observed across 19 m…
-
TokenGate module accelerates LLM prefill for long contexts
Researchers have developed TokenGate, a novel module designed to significantly accelerate the prefill process in large language models (LLMs) for long-context inputs. This lightweight component enables LLMs to selective…
-
TRACE framework models clinical trajectories for improved inpatient risk prediction
Researchers have developed TRACE, a new framework for predicting inpatient risks by modeling clinical events as temporal trajectories. This approach organizes daily clinical data, including electronic health records, ch…
-
FocusGUI adapter boosts GUI agent accuracy by prioritizing multimodal evidence
Researchers have introduced FocusGUI, a novel adapter designed to enhance the performance of graphical user interface (GUI) agents. This system dynamically selects relevant multimodal evidence, such as screenshots and a…
-
TokenWorth framework learns valuable memories for resource-constrained lifelong agents
Researchers have introduced TokenWorth, a novel memory framework designed for lifelong agents that operate under strict resource limitations. This framework learns to identify which past experiences are most valuable to…
-
Stateful Harness framework improves interactive world models
Researchers have introduced Stateful Harness, a novel framework designed to enhance the consistency and rule-following capabilities of interactive world models. This system decouples the process of updating the world st…
-
Hugging Face researchers unveil HOLI-WORLD for egocentric AI learning
Researchers have introduced HOLI-WORLD, an integrated ecosystem designed to automate the creation of large-scale learning resources from egocentric recordings. This system aims to enhance machine understanding of dynami…
-
BiDex-100 benchmark launched for bimanual dexterous manipulation tasks
Researchers have introduced BiDex-100, a new simulation benchmark designed to evaluate vision-language-action models in bimanual dexterous manipulation tasks. The benchmark encompasses over 100 diverse simulation tasks,…
-
FineSplat advances 3D Gaussian Splatting with fine-grained patches and dual-grouping
Researchers have introduced FineSplat, a novel feed-forward 3D Gaussian Splatting framework designed to improve the recovery of fine-grained geometry and appearance in 3D scenes. The method utilizes fine-grained image p…
-
New TCoVP method enhances video reasoning segmentation with grounded evidence
Researchers have introduced Temporal Chain of Video Primitives (TCoVP), a novel approach to video reasoning segmentation (VRS). TCoVP addresses the gap in text-only Chain-of-Thought methods by explicitly linking entitie…
-
Progress-Aware Distillation tackles stagnation in small language model agents
Researchers have developed a new technique called Progress-Aware Distillation (ProD) to address stagnation issues in small language model (SLM) agents. This method detects and penalizes repetitive, unproductive loops in…