All published AI news and analysis.

Researchers propose Prediction-Powered Smoothing (PP-S) and Prediction-Powered Taxonomy Smoothing (PP-TS) for more accurate AI system evaluation across diverse domains. These methods build on small area estimation techniques to improve point and interval estimates, especially when domain-specific data is limited.

Researchers have introduced ScienceIDE, an infrastructure designed to transform scientific code repositories into programmable environments for AI agents. This system aims to overcome the 'scientific experience bottleneck' by enabling agents to generate, execute, and verify scientific tasks, facilitating agent learning and scientific practice.

Researchers have introduced PANORAMA, a vision-language model designed for panoptic grounded captioning, which aims to generate detailed scene descriptions with precise pixel-level grounding. The model is accompanied by PanoCaps, a new human-annotated benchmark dataset for training and evaluating this task.

A new research paper introduces a method using difference of means (DoM) vectors derived from internal model representations to detect reward hacking in open-source LLMs. This white-box approach offers a cost-effective alternative to traditional LLM monitors, showing comparable effectiveness and the ability to discover novel hacking behaviors.

Researchers introduce PointZero, a novel pre-training objective called 3D point track completion for learning transferable 3D dynamics without requiring robot action labels. This method utilizes a diverse dataset of 2.9 million synthetic frames and a transformer architecture to predict future 3D tracks of observed points, outperforming prior methods and demonstrating utility in downstream tasks like action-conditioned dynamics prediction and imitation learning.

OpenAI's latest article details how ChatGPT Work and Codex analytics can help teams monitor AI usage and expenses. The tools aim to identify skill gaps and demonstrate how AI adoption contributes to business objectives.

OpenAI has introduced a new framework for tracking, investigating, and disclosing instances of model misalignment. This initiative is accompanied by six initial reports detailing unexpected or concerning model behaviors.

OpenAI has published economic research detailing how workers are integrating AI into their roles beyond conventional applications. The study identifies emerging activities that are becoming routine parts of professional workflows.

OpenAI is partnering with AARP to offer free, in-person ChatGPT workshops for 1,000 older adults across 10 U.S. cities. The initiative aims to equip participants with practical AI skills in a safe environment.

OpenAI is launching new AI-powered advertising experiences, including Sponsored Agents and tools designed for marketers. These new offerings will feature integrations with platforms like HubSpot and Shopify.