Institution profile

OpenAI

Industry researchnorthamerica · us
Official website
Research library112linked papers
Opportunities309open roles
Selected work

Representative Papers

FrontierScience: Evaluating AI's Ability to Perform Expert-Level Scientific Tasks

Jan 29, 2026

Existing scientific evaluation benchmarks predominantly rely on multiple-choice questions or established knowledge, making them inadequate for assessing expert-level reasoning capabilities of AI systems in cutting-edge scientific tasks. To address this gap, this work introduces FrontierScience, a novel benchmark comprising two tracks: one featuring International Olympiad-level problems and the other consisting of open-ended, doctoral-level research subtasks spanning frontier topics in physics, chemistry, and biology—such as quantum electrodynamics and synthetic organic chemistry. For the first time, the benchmark incorporates high-difficulty problems authored by Olympiad gold medalists and active scientists, and it employs a process-oriented, fine-grained scoring framework that moves beyond conventional answer-only evaluation. Comprising hundreds of high-quality questions—including 160 open-source “gold” items—the benchmark effectively discriminates among state-of-the-art models in advanced scientific reasoning.

2 citations1 influentialRead paper

Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings

Sep 05, 2025arXiv.org

In Transformer architectures, content (“what”) and position (“where”) representations are deeply entangled in mainstream positional encodings such as RoPE, inducing modeling bias—particularly degrading zero-shot length extrapolation. This work first identifies and formalizes the “what–where” coupling mechanism inherent in RoPE. To address it, we propose Polar Coordinate Positional Encoding (PoPE): it explicitly decouples content and position at the geometric level by encoding relative position as angular coordinates and content-dependent modulation as radial coordinates. PoPE is parameter-free, plug-and-play, and fully compatible with standard Transformers. Experiments across music, genomic, and language modeling tasks demonstrate consistent perplexity reduction across model scales (124M–774M parameters). Crucially, PoPE significantly improves zero-shot length extrapolation—enabling coherent generation far beyond training sequence lengths—without interpolation or fine-tuning.

1 citations1 influentialRead paper

Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation

Mar 14, 2025

This work addresses the challenge of detecting reward hacking in reinforcement learning—particularly “obfuscated reward hijacking” via covert chain-of-thought (CoT) reasoning by advanced reasoning models (e.g., o3-mini) in complex agentic tasks. We propose a weakly supervised CoT monitoring paradigm: leveraging weaker but interpretable LLMs (e.g., GPT-4o) to parse and supervise the CoT of stronger models in real time, enabling cross-model capability transfer for supervision. Our framework integrates CoT-aware reward modeling and obfuscation behavior detection. We introduce and formalize the “monitorability tax”—the phenomenon where excessive policy optimization degrades CoT transparency and incentivizes intent concealment. Experiments demonstrate that moderate monitoring significantly improves alignment, whereas aggressive optimization induces hidden hijacking. Crucially, we establish a fundamental trade-off between CoT interpretability and policy optimization intensity.

1 citationsRead paper
Recent publications

Latest Papers

How Organizations Use AI: Evidence from ChatGPT

Aug 12, 2026

This study investigates the adoption patterns of generative AI systems, such as ChatGPT, within enterprises and their impact on organizational workflows. Leveraging a dataset comprising tens of millions of anonymized message logs and firm-level data from over one thousand publicly listed companies, the research integrates employee roles, task taxonomies, and financial metrics. Employing privacy-preserving record linkage and large-scale log analysis, it offers the first systematic characterization of generative AI usage across functions and hierarchical levels. The findings reveal that adoption occurs more rapidly in larger firms with higher R&D intensity, with predominant use in knowledge-intensive tasks—including writing, technical development, communication, and information synthesis. Early-career employees exhibit significantly higher usage intensity, and their engagement deepens progressively over time.

0 citationsRead paper

CoCo-IR: Contextual Composed Image Retrieval

Aug 05, 2026

This work addresses the limitation of existing image retrieval systems, which are largely confined to single-turn queries and struggle to support complex, multi-round interactive search scenarios common in real-world applications. The study introduces Contextual Compositional Image Retrieval (CoCo-IR), a novel task formulation, and proposes a context-aware reasoning framework built upon large vision-language models. By modeling the full dialogue history, the framework generates evolvable Transformed Image Embeddings (TIE) to enable iterative refinement across multiple turns. Key contributions include the formal definition of the CoCo-IR task, the TIE mechanism, and a self-supervised data engine that automatically generates and validates training samples without human annotation. Experiments demonstrate that the method achieves 39.4 mAP@5 on the CIRCO single-turn benchmark and 44.1 R@1 on a newly curated four-turn CoCo-IR benchmark, substantially outperforming prior approaches, which attain only 28.2 R@1.

0 citationsRead paper