Deniz Qian

Master’s student in Computer Science at NYU Courant.

[Deniz Qian @denq1an]

🚨NEW PAPER🚨
How can we comprehensively retrieve all relevant docs for multi-answer QA? Agentic search doesn't help.

Introducing RVR, an iterative framework that conditions on prior docs to maximize answer coverage.

📈10% answer recall gain on QAMPARI
w/@hungting_chen @eunsolc

[Deniz Qian @denq1an]

Very excited to have been part of Cohere’s newest coding model launch! 🚀

[Cohere @cohere]

Introducing Cohere's first open-source coding model: North Mini Code

Small & efficient, designed for agentic performance and built for community input.

[Yuhan Liu @YuhanLiu_nlp]

Can LLMs generate diverse outputs for open-ended questions? Is it helpful if we ensemble outputs from multiple models? We study 18 LLMs on 4 datasets and find that no single model is best at generating diverse outputs 👇/ 🧵

[Hung-Ting Chen @hungting_chen]

It was nice collaborating with @denq1an on this project!

Comprehensively retrieving all answers is hard for agentic approaches that are optimized to cover a single answer.

We propose a iterative framework that uncovers new answers based on previously retrieved information.

[Sumit @_reachsumit]

RVR: Retrieve-Verify-Retrieve for Comprehensive Question Answering

@denq1an et al. use an LLM verifier to identify high-quality documents and conditions subsequent retrieval on verified results to maximize answer coverage.

📝 https://t.co/WtoqzHQqLJ
👨🏽‍💻 https://t.co/mg4b7UXDZg

[Deniz Qian @denq1an]

Takeaway: Instead of making LLM agents more complex, we adapt the retriever to the inference setting.

More analysis in the paper!
Paper: https://t.co/tu4lH4Typa
Happy to discuss!

[Deniz Qian @denq1an]

🚨NEW PAPER🚨
How can we comprehensively retrieve all relevant docs for multi-answer QA? Agentic search doesn't help.

Introducing RVR, an iterative framework that conditions on prior docs to maximize answer coverage.

📈10% answer recall gain on QAMPARI
w/@hungting_chen @eunsolc

[Deniz Qian @denq1an]

Multi-turn retrieval consistently improves coverage. Each additional round recovers new gold documents and new answers. Even with just 2 rounds, we see substantial gains. With stronger verification, improvements continue across turns.

[Deniz Qian @denq1an]

@aidywavy wrapped