Jonathan Tonglet
@tongletj.bsky.social
🇧🇪 ELLIS PhD student @ TUDarmstadt and KULeuven
Working on multimodal fact-checking
https://jtonglet.github.io
Working on multimodal fact-checking
https://jtonglet.github.io
Reposted by Jonathan Tonglet
UKP is at #ACL2025 in Vienna!
Meet @igurevych.bsky.social, our postdocs, and our PhD students, and be sure to check out their posters 🎉
More information about our ACL and TACL papers below ⬇️
Meet @igurevych.bsky.social, our postdocs, and our PhD students, and be sure to check out their posters 🎉
More information about our ACL and TACL papers below ⬇️
July 28, 2025 at 3:12 PM
UKP is at #ACL2025 in Vienna!
Meet @igurevych.bsky.social, our postdocs, and our PhD students, and be sure to check out their posters 🎉
More information about our ACL and TACL papers below ⬇️
Meet @igurevych.bsky.social, our postdocs, and our PhD students, and be sure to check out their posters 🎉
More information about our ACL and TACL papers below ⬇️
Extended preprint with new experiments arxiv.org/abs/2502.20503 🚨
We identify a key vulnerability of multimodal LLMs: they are easily deceived by misleading charts🫣
This vulnerability opens many doors for future work in automated chart understanding, be sure to check it out!
We identify a key vulnerability of multimodal LLMs: they are easily deceived by misleading charts🫣
This vulnerability opens many doors for future work in automated chart understanding, be sure to check it out!
MLLMs can easily be misled by manipulated charts to produce false answers⚠️📊
We show the extent of this vulnerability and how to fix it by modifying the misleading chart image or replacing it by another modality - 🧵(1/5)
📄Paper: arxiv.org/abs/2502.20503
💻Project: ukplab.github.io/arxiv2025-mi...
We show the extent of this vulnerability and how to fix it by modifying the misleading chart image or replacing it by another modality - 🧵(1/5)
📄Paper: arxiv.org/abs/2502.20503
💻Project: ukplab.github.io/arxiv2025-mi...
April 30, 2025 at 6:07 PM
Extended preprint with new experiments arxiv.org/abs/2502.20503 🚨
We identify a key vulnerability of multimodal LLMs: they are easily deceived by misleading charts🫣
This vulnerability opens many doors for future work in automated chart understanding, be sure to check it out!
We identify a key vulnerability of multimodal LLMs: they are easily deceived by misleading charts🫣
This vulnerability opens many doors for future work in automated chart understanding, be sure to check it out!
Very excited to share this work which opens new research directions for chart reasoning with LLMs!
We find that, like humans, multimodal LLMs are vulnerable to misleading charts, which severely impairs their chart QA abilities.
Check out our preprint, data, and code - more information ⬇️
We find that, like humans, multimodal LLMs are vulnerable to misleading charts, which severely impairs their chart QA abilities.
Check out our preprint, data, and code - more information ⬇️
MLLMs can easily be misled by manipulated charts to produce false answers⚠️📊
We show the extent of this vulnerability and how to fix it by modifying the misleading chart image or replacing it by another modality - 🧵(1/5)
📄Paper: arxiv.org/abs/2502.20503
💻Project: ukplab.github.io/arxiv2025-mi...
We show the extent of this vulnerability and how to fix it by modifying the misleading chart image or replacing it by another modality - 🧵(1/5)
📄Paper: arxiv.org/abs/2502.20503
💻Project: ukplab.github.io/arxiv2025-mi...
March 10, 2025 at 8:46 AM
Very excited to share this work which opens new research directions for chart reasoning with LLMs!
We find that, like humans, multimodal LLMs are vulnerable to misleading charts, which severely impairs their chart QA abilities.
Check out our preprint, data, and code - more information ⬇️
We find that, like humans, multimodal LLMs are vulnerable to misleading charts, which severely impairs their chart QA abilities.
Check out our preprint, data, and code - more information ⬇️
Reposted by Jonathan Tonglet
One more paper at #NAACL:
»COVE: COntext and VEracity prediction for out-of-context images« by Jonathan Tonglet (@tongletj.bsky.social), Gabriel Thiem (@gthiem.bsky.social) & Iryna Gurevych (@igurevych.bsky.social)
»COVE: COntext and VEracity prediction for out-of-context images« by Jonathan Tonglet (@tongletj.bsky.social), Gabriel Thiem (@gthiem.bsky.social) & Iryna Gurevych (@igurevych.bsky.social)
February 1, 2025 at 9:31 AM
One more paper at #NAACL:
»COVE: COntext and VEracity prediction for out-of-context images« by Jonathan Tonglet (@tongletj.bsky.social), Gabriel Thiem (@gthiem.bsky.social) & Iryna Gurevych (@igurevych.bsky.social)
»COVE: COntext and VEracity prediction for out-of-context images« by Jonathan Tonglet (@tongletj.bsky.social), Gabriel Thiem (@gthiem.bsky.social) & Iryna Gurevych (@igurevych.bsky.social)
Reposted by Jonathan Tonglet
Reposted by Jonathan Tonglet
Sorry that I’m missing a lot of people. If you’re working on making NLP models more culturally aware, please DM me to be added.
go.bsky.app/tRMpng
go.bsky.app/tRMpng
November 21, 2024 at 8:23 PM
Sorry that I’m missing a lot of people. If you’re working on making NLP models more culturally aware, please DM me to be added.
go.bsky.app/tRMpng
go.bsky.app/tRMpng