Emilia Góral

[Gracjan Góral @gracjan_goral]

industry vs. academia compute at @CVPR

[ML in PL @MLinPL]

We went through CVPR 2026 and pulled out every paper with a Polish author. One pattern jumped out immediately: more than a third are 3D. Novel view synthesis, Gaussian splatting, feed-forward reconstruction. Gaussians everywhere.

[Lute @lutecompany]

Can Vision Language Models truly make robots see the world from a human perspective?
Our teammate - @gracjan_goral is exploring exactly that at #CVPR2026 🤖
Join the session and listen about "Beyond Recognition: Evaluating Visual Perspective Taking in Vision Language Models".

[Gracjan Góral @gracjan_goral]

met some great new people. answered all the questions (i hope) -- some were really tough. thanks to all my collaborators: @ZiarkoAlicja @mic_nau @maciejwolczyk @PiotrRMilos @michalkosinski and @lutecompany!

[Michal Tyrolski @mtyrolski]

🚀 We just released our open-source codebase and a public leaderboard for evaluating reasoning and progress sense of AI models! Big things are coming - don’t miss out! 👀
thanks to:
@pfbudzianowski, @gracjan_goral, @wisnios_emilia, @ihorbeaver, @viktor_vrp and @KrzysztofTWalas

[Paweł Budzianowski @pfbudzianowski]

We are happy to share our v1 version of the temporal progress leaderboard (https://t.co/wD6s9uhGEf). Visual temporal reasoning is an essential characteristic that VLM/VLA should exhibit natively for proper data curation and adaptation for robotics tasks.

[Gracjan Góral @gracjan_goral]

now, you can send us your vlm or dataset, and we'll tell you how good it is. seriously, send it.

thnaks to: @pfbudzianowski, @wisnios_emilia, @ihorbeaver, @viktor_vrp, @mtyrolski and @KrzysztofTWalas

[Sebastian Dziadzio @sbdzdz]

If you're in Vienna for ACL, @adhiraj_ghosh98 and I will be presenting our work on benchmarking language and vision-language models in the post-dataset era on Wednesday from 11 am in Hall 4/5. Come say hi!

Project page 🔗
https://t.co/1SVlmJZ27S

[Ahmad Beirami @abeirami]

Instead of complaining that peer review is dead, take a positive step to improve it today. The reviewers are not aliens, they are us!

- Revise your review and make it clear. Identify the crucial points that impacted your score negatively and positively.

- If the paper is lacking information about its claims, communicate your asks and the reasoning concretely. Don't just ask for 2 more experiments because you feel the authors didn't work hard enough. Don't ask for experiments unless they verify a hypothesis (which you clearly explained).

- Look for the missing information that you identified and make sure they are not in the paper.

- Recommend acceptance if the paper's claim is adding a new nugget of information to the literature (no matter the size of the nugget), and if the paper has substantiated the claim via theoretical / empirical evidence.

Believe me, this doesn't take much time, and will improve the state of peer review significantly!

[Paweł Budzianowski @pfbudzianowski]

I'm so happy when VLM spatial folks look into robotics to find new pre-training tasks! @gracjan_goral @Emilia Wiśnios did awesome job bringing up generative task completion (from @JasonMa2020 and co.) with open-source models. 1/2

[Emilia Góral @emigoral]

@ziv_ravid @pfbudzianowski

[ML in PL @MLinPL]

🚨 One week left to submit contributed talks, posters, and tutorials to the ML in PL Conference 2025!

Don't miss your chance to present at this year's conference:
👉 Main Conference talks and posters (October 15-17, 2025)
👉 Tutorials (October 18, 2025)

✓ Talk and poster presenting authors receive free conference tickets
✓ Tutorial presenters receive 2 free conference tickets

Submit now:
👉 Call for Contributions: https://t.co/NuDliitZBF
👉 Call for Tutorials: https://t.co/oaV023qc3b

Find more information here:
👉 Call for Contributions: https://t.co/C7k2ObGnQz
👉 Call for Tutorials: https://t.co/0mSi5fX5ZN

[Emilia Góral @emigoral]

Check out both of my papers:

- Wait, that's not an option: LLMs Robustness with Incorrect Multiple-Choice Options (https://t.co/837C7dJn6q)
- Behind Closed Words: Creating and Investigating the forePLay Annotated Dataset for Polish Erotic Discourse (https://t.co/UzTSqeiADY)

[Emilia Góral @emigoral]

Presenting two posters at the same poster session with a sprained ankle might be challenging, but I hope I'll manage. See you on Monday, July 28, at 11 a.m. I will be the one with the orthopaedic crutch 😂

[Gracjan Góral @gracjan_goral]

what if instructions didn't just tell you what to do, but validated your inner reasoning?

with @wisnios_emilia, @piotrsankowski, and @pfbudzianowski, we explored this idea. excited to present our work at #ACL2025 in Vienna next week!

paper: https://t.co/A8k3wRlpcp

[Gracjan Góral @gracjan_goral]

so some time ago we asked: what’s the threshold between following instructions and accuracy in llms; like:

what is the capital of austria? reply only a or b;

a. an apple
b. euler

we know the answer and will share it at #ACL2025

check it out: https://t.co/A8k3wRlpcp

[Piotr Sankowski @piotrsankowski]

W informatyce bardzo lubimy konferencje oznaczać skrótami. A ACL to coroczne spotkanie stowarzyszenia lingwistyki obliczeniowej (Annual Meeting of the Association for Computational Linguistics) i jest to jedna z konferencji o randze A*.

Miło mi poinformować, że nasza praca dotycząca "refleksyjności" LLMów została zaakceptowana na ACL 2025: https://t.co/dDXV2he5Nh. Dotyczy ona wyzwań związanych z podejmowaniem decyzji przez duże modele językowe (LLM) w sytuacjach, które wymagają równowagi między wnioskowaniem a wiernością instrukcjom. Badamy, czy modele te są skłonne bardziej podążać za poleceniami niż za logiką i prawdą, zwłaszcza gdy polecenia są mylące. Wprowadzamy nową metrykę – refleksyjny osąd – która pozwala lepiej zrozumieć, jak różne techniki dostrajania wpływają na zdolność modeli (oraz ludzi) do krytycznej oceny poprawności instrukcji.

[Emilia Góral @emigoral]

@mar_kar_ @jennajrussell @shanearora0 Congrats! Hope to see you in Vienna!

[Emilia Góral @emigoral]

So I guess #ACL results are out 🥳

[Gracjan Góral @gracjan_goral]

wow guys; in 3 days, our perspective datasets have over 200 downloads from hf (in total); i know; that isn't much; but it is honest work; thanks!

datasets: https://t.co/CGFSnAyuKx

let me know if you want another blender-perspective dataset to test VLMs and humans;

[Gracjan Góral @gracjan_goral]

people need to take perspective (emotional and geometrical); robots too (I think); emotional is hard to measure (at least for me lol); geometrical slightly easier;