←
Desmond Elliott3,408 posts
Desmond Elliott avatar

Desmond Elliott

@delliott

Associate Professor at the University of Copenhagen. I work on Vision-Language and Tokenization-free NLP.

◷ Joined August 2007▧ 195 Photos and videos
Chris Dyer avatar
⟳ Desmond Elliott retweeted
Chris Dyer@redpony·Nov 16

My team at Google DeepMind is hiring a student researcher (NYC, 2027) to work on how we evaluate sign language AI. Deaf and hard of hearing researchers are especially encouraged to apply, as are fluent signers. Get in touch: forms.gle/Gwfmp1cKGQ8oN2sSA

Sebastian Schuster avatar
⟳ Desmond Elliott retweeted
Sebastian Schuster@sebschu·Nov 13

Come be my colleague in Vienna! The Vienna Science and Technology Fund is again funding up to four "Vienna Research Groups". They come with up to EUR 1.8 Million in funding, a tenure-track position with a lightweight tenure process and lots of other benefits.

Sunnie S. Y. Kim ☀️ avatar
⟳ Desmond Elliott retweeted
Sunnie S. Y. Kim ☀️@sunniesuhyoung·Nov 12

How are LLMs changing scientific peer review, and what happens when conferences implement different LLM-use policies for reviewers? In our new paper, we investigate these questions through a large-scale RCT and an anonymous survey at #ICML2026. 1/5

Post attachment
Desmond Elliott avatar
Desmond Elliott@delliott·Oct 28

Excited about this new project with @monurcan55, where he shows that omnimodal models can self-improve by generating their own data 📈

Can a vision-language model fix its own mistakes at test time, without a single label? With SIMIT, the model imagines practice problems before answering. It decides each answer first, then creates an image that makes it true. +7.20% on 17 benchmarks 🧵 monurcan.github.io/simit/

0:38
Manos Zaranis avatar
⟳ Desmond Elliott retweeted
Manos Zaranis@ManosZaranis·Oct 22

🎬 How Well Do Vision-Language Models Understand Movies? Introducing MF²: Movie Facts and Fibs, accepted at #EMNLP2026 in Budapest 🇭🇺🎉 A benchmark for video narrative comprehension, not just shallow recall 🧵👇

Post attachment
Niloofar avatar
⟳ Desmond Elliott retweeted
Niloofar@niloofar_mire·Oct 21

Super excited to be giving a talk titled 'what goes where?' at the Copenhagen NLP Symposium tomorrow! I'll be sharing cool research we have been doing on long horizon agentic tasks involving privacy and chemistry @LTIatCMU, @CMUEngineering and will also discuss our most recent works involving long-horizon simulations and user modeling @humansand! also will be recruiting for both, come chat!

Post attachmentPost attachmentPost attachmentPost attachment
Desmond Elliott avatar
Desmond Elliott@delliott·Oct 6

In Rate-Utility Frontiers for Language Encodings: Comparing Tokens, Bytes, and Pixels Under Controlled Linguistic Content, @IngoZiegler shows that the "best" input encoding depends on the downstream task in carefully controlled experiments. Paper: arxiv.org/abs/2607.16117

Desmond Elliott avatar
Desmond Elliott@delliott·Oct 2

📣 I'm hiring a Ph.D student through ELLIS to join my group at the University of Copenhagen. We work on tokenization-free language models (think: bytes, characters, and pixels). Your office will be at the Østervold Observatory in the beautiful Copenhagen Botanical Gardens.

Post attachment
Marco🍞 avatar
⟳ Desmond Elliott retweeted
Marco🍞@marco_computers·Sep 30

🚨 [Token][ization] Paper Alert 🚨 Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!

Post attachment