Twitxy Saved

Stanford OVAL

@StanfordOVAL

A research lab developing Expert AI, training large language models to prevent hallucination and enable knowledge-oriented, multilingual and multimodal tasks.
Stanford, CA oval.cs.stanford.edu Joined October 2018
259 Following 1.8K Followers 245 Posts

Latest posts 5

Stanford OVAL @StanfordOVAL
Meet SHELDON, our new scientific reasoning harness that helps AI agents explore observations and hypotheses together and pursue discoveries beyond their initial guesses.
Harshit Joshi Verified account @harshitj__ Oct 15
Scientific 🔬 progress often starts with an unexpected observation that sparks a new hypothesis.

But AI agents often start with a hypothesis, then look for observations to support it.

We believe agents should actively explore both observations and hypotheses. When observations are only a byproduct of hypothesis testing, we risk missing discoveries we weren’t looking for.

⚡ Meet SHELDON: a scientific reasoning harness that models separate hypothesis and observation spaces and generalizes across medical diagnosis, system-log investigations, ARC-AGI 2, and even autoresearch.

Bazinga!
View quoted post on X
Stanford OVAL @StanfordOVAL
Check out Cyrus' new work just accepted to COLM 2026!
Cyrus (Zikai) Zhou Verified account @cyruszzhou Jul 8
Excited to share that SatIR has been accepted to COLM 2026! 🎉 Grateful to my incredible collaborators across Stanford and Mayo Clinic. The core idea: for high-stakes retrieval, don't just ask "does this sound relevant?" - ask "is this actually possible?"

Also accepted - DataSTORM from @StanfordOVAL led by @ShichengGLiu , an LLM agent that runs deep research over massive structured databases, turning raw data into coherent analytical narratives (and beating ChatGPT Deep Research): arxiv.org/abs/2604.06474
View quoted post on X
Stanford OVAL @StanfordOVAL
Check out our new work on vision-language models for OCR on historical documents!
Sina Semnani Verified account @sina_semnani Nov 18, 2025
There are many new VLMs/OCR models out there, but most still fail on historical documents.

We’re releasing CHURRO, a 3B open-weight vision-language model that outperforms Gemini 2.5 Pro on historical OCR while being 15.5× more cost-effective.

From our EMNLP 2025 (Main) paper:
View quoted post on X