ICML 2025
PulseAugur coverage of ICML 2025 — every cluster mentioning ICML 2025 across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
AI models favor similar code, risking overlooked bugs in reviews
A new study suggests that AI models tend to favor and agree with outputs from similar models, potentially leading to overlooked errors in code reviews. Researchers found that AI judges often scored models more favorably…
-
RAG systems need better retrieval, not just longer context, guide says
A 2026 guide to Retrieval-Augmented Generation (RAG) systems highlights that failures often stem from the retrieval process rather than the language model itself. Despite advancements in long-context windows, retrieval …
-
New benchmark AlphaIdeaBench evaluates AI research idea generation
Researchers have introduced AlphaIdeaBench, a new benchmark designed to evaluate AI agents that generate research ideas under specific resource constraints. The benchmark includes 100 ideation tasks, each with a defined…
-
New detection method uses physiological signals to identify talking-face deepfakes
Researchers have developed a new method for detecting talking-face deepfakes by analyzing physiological signals, specifically remote photoplethysmography (rPPG) waveforms. Their framework, utilizing a model called Rhyth…
-
New benchmark evaluates LLMs' research planning capabilities
Researchers have developed a new benchmark, Idea2Plan, to evaluate the capability of large language models (LLMs) in transforming conceptual research ideas into structured plans. This benchmark, derived from ICML 2025 a…
-
New research details minimax regret for bilateral trade with infinite variance
A new research paper published on arXiv introduces a novel algorithm for bilateral trade under heavy-tailed valuations, specifically addressing scenarios where trader valuations exhibit infinite variance. The proposed m…
-
Google Research enhances time-series forecasting with few-shot learning
Google Research has developed a new method to enhance time-series forecasting models by enabling them to learn from a few examples at inference time. This approach, called In-Context Fine-Tuning (ICF), builds upon their…
-
Google Research unveils CTCL for privacy-preserving synthetic data generation
Google Research has developed a new privacy-preserving synthetic data generation algorithm called CTCL, designed for resource-constrained AI applications. Unlike previous methods that require fine-tuning large language …
-
Which Agent Causes Task Failures and When?Researchers from PSU and Duke explores automated failure attribution of LLM Multi-Agent Systems
Researchers from Penn State University and Duke University, alongside collaborators from institutions including Google DeepMind, have introduced a new research problem called "Automated Failure Attribution" for LLM Mult…
-
Google unveils Simula and CTCL for advanced synthetic data generation
Google Research has introduced Simula, a framework that treats synthetic data generation as a mechanism design problem. This approach allows for fine-grained control over dataset characteristics like coverage, complexit…