Can LLMs support behavioral health documentation without compromising quality?
Our ACL 2025 paper introduces TN-Eval: a comprehensive framework for evaluating therapy notes, co-designed with 22 licensed therapists.
Key finding: In blind tests, therapists rated LLM-generated notes as superior to human-written ones! ๐ค
With Lei Xu @XuLeonard, Qianchu Liu @qianchul, Jon Burnsky, Drew Bertagnolli @onemedical, Chaitanya Shivade @ekshoonyame@AmazonScience@ICatGT #ACL2025 #ClinicalNLP
๐ Dataset: https://t.co/ksELdNLKwD
๐ Code: https://t.co/H8HdXv24Nc
๐ Paper: https://t.co/fqNpigcg4E
This aligns with the growing trend โ echoed by efforts like OpenAI's #HealthBench โ of using structured criteria to guide and evaluate LLM applications in clinical settings.
1/ ๐ Excited to share our latest work โ accepted to #ICLR2025 and featured in MIT News today!
๐๏ธ MIT News: https://t.co/lPbVmAbf9i
๐ Paper: https://t.co/AoBsV3qeUo
๐งต Thread ๐
Really excited to announce that our NeurIPS 2019 paper on 'Modeling tabular data using conditional GAN' has surpassed 1k citations! It's inspiring to see researchers applying the model innovatively in the era of LLMs. #NeurIPS#GAN#SyntheticData#MIT
Researchers w/ MIT's Laboratory for Information & Decision Systems in the College of Computing released the Synthetic Data Vault โ a one-stop shop where users can get as much data as they need for their projects, without compromising privacy.
https://t.co/I1MZzoql0F