Eric

PhD student at @Cambridge_CL, researching NLP with @vlachos_nlp
Member of @Kings_College

[Eric @Eric_chamoun]

“I want to build X.”
What needs to happen first?

Behind every scientific contribution is a pathway of enabling methods, datasets, tools, models, ideas, and prior work.

We introduce SciPaths to study whether LLMs can predict these pathways. 🧬🧭

[Eric @Eric_chamoun]

Poster session is underway! Come find us at board 500 🙌🏻 #ICML2026

[Eric @Eric_chamoun]

Excited to be in Seoul for ICML! I will be presenting our work on forecasting scientific discovery pathways on Saturday July 11th at the:

- AI Forecasting Workshop: 12:15-13:30
- AI4Science Workshop: 15:55 - 16:55

Would also love to catch up with anyone who’s there!

[Eric @Eric_chamoun]

Excited to be in Seoul for ICML! I will be presenting our work on forecasting scientific discovery pathways on Saturday July 11th at the:

- AI Forecasting Workshop: 12:15-13:30
- AI4Science Workshop: 15:55 - 16:55

Would also love to catch up with anyone who’s there!

[Eric @Eric_chamoun]

“I want to build X.”
What needs to happen first?

Behind every scientific contribution is a pathway of enabling methods, datasets, tools, models, ideas, and prior work.

We introduce SciPaths to study whether LLMs can predict these pathways. 🧬🧭

[Yizhou Chi @YizhouChi]

WorldReasoner is now open-source! Dive into the code and benchmark here:
GitHub: https://t.co/a8Qu4Glhov

Thanks to @Eric_chamoun @ZifengDing6 @vlachos_nlp
(4/4)

[Yizhou Chi @YizhouChi]

How to test agent forecasting without data contamination?

Isolate them in a temporal sandbox⏳

Introducing WorldReasoner: a new framework that restricts agents to simulated historical dates to test genuine reasoning over time-bounded evidence

https://t.co/vi1iN6BrpX (1/4)

[Eric @Eric_chamoun]

What reusable contributions did your paper create & what made them possible? 🧪

Try the demo with an arXiv paper w/ downstream citations. It finds target contributions from downstream use, then their enabling contributions + prior-work groundings.

Demo: https://t.co/jfSwn3UcIp

[Eric @Eric_chamoun]

“I want to build X.”
What needs to happen first?

Behind every scientific contribution is a pathway of enabling methods, datasets, tools, models, ideas, and prior work.

We introduce SciPaths to study whether LLMs can predict these pathways. 🧬🧭

[Eric @Eric_chamoun]

This matters for AI4Science agents 🤖🔬

A system that proposes a research direction should also reason about feasibility:

• what must already exist
• which prior work provides it, if any
• what needs to be built next

That is the capability SciPaths is designed to test.

[Eric @Eric_chamoun]

“I want to build X.”
What needs to happen first?

Behind every scientific contribution is a pathway of enabling methods, datasets, tools, models, ideas, and prior work.

We introduce SciPaths to study whether LLMs can predict these pathways. 🧬🧭

[Eric @Eric_chamoun]

Task B shows why decomposition matters 🔎

With gold enabling contributions, prior-work grounding improves substantially.

But model-generated decompositions do not yet improve end-to-end grounding over direct claim retrieval.

Often, the bottleneck is knowing what to search for.

[Eric @Eric_chamoun]

📄 Paper: https://t.co/xlAIz77Kpd
💾 Code & data: https://t.co/fZ8YM9mdxQ

[Eric @Eric_chamoun]

Excited to share that our paper on automatically uncovering how NLP papers frame their research will be presented this week at the EMNLP 2025 main conference in Suzhou by @vlachos_nlp! 🎉

Paper, code and data below👇

@michael_sejr @nedjmaou

[Eric @Eric_chamoun]

What are NLP papers really saying about the purpose and use of their models/datasets? 🤔

Who are they for? What problems do they solve? How are they used?

We built a framework + tool to:
(1) analyze framing trends across papers
(2) help authors reflect on their own framing 🧵

[Matthieu @matt_tehenan]

Just released a new paper !

We show that LLMs encode the relative position of objects in a 3D subspace in their weights.

A rudimentary world model.

(🧵)

[Nedjma Ousidhoum نجمة أوسيدهم @nedjmaou]

Check out our new preprint on analysing the research framings of #NLProc papers (w/ a focus on automated fact-checking, hate speech and related tasks)
With fantastic collaborators @Eric_chamoun (lead) @michael_sejr and @vlachos_nlp !
📝👇🏻