Records found: 7
'Anthropic open-sourced Petri, a tool that automates alignment audits by orchestrating auditor and judge agents to probe LLMs across multi-turn, tool-augmented scenarios, revealing misaligned behaviors in a 14-model pilot.'
'A practical tutorial showing how to combine LangChain tools with XGBoost to create an interactive end-to-end ML pipeline that generates data, trains a model, evaluates performance, and visualizes insights.'
'A hands-on tutorial showing how to create synthetic multi-series data, run several GluonTS estimators conditionally, evaluate probabilistic forecasts, and build advanced visualizations for model comparison.'
'Step-by-step guide to testing OpenAI models with deepteam single-turn attacks, including prompt injection, leetspeak, Base64, ROT13 and multilingual tests.'
TabArena offers a dynamic, community-driven benchmarking platform for tabular machine learning, emphasizing reproducibility, ensembling, and extensive hyperparameter tuning to deliver state-of-the-art performance insights.
Meta and collaborators developed a novel framework to accurately quantify how much language models memorize training data, estimating GPT models store around 3.6 bits per parameter and providing new insights into memorization versus generalization.
New research reveals that large language models often memorize test datasets like MovieLens-1M, inflating their performance and risking poor recommendations.