HKUST NLP (@hkustNLP) header
HKUST NLP (@hkustNLP) profile picture

HKUST Natural Language Processing Research #NLProc

Joined October 2024
Photo by HKUST NLP (@hkustNLP)
Great work! @ZhaoweiWang4
MMLongBench Benchmarking Long-Context Vision-Language Models Effectively and Thoroughly
2
784
Check out our new work on multimodal long-context!
๐Ÿš€ We release MMLongBench: Benchmark for evaluating long-context VLMs. ๐Ÿ“Š 13,331 examples across 5 tasks: โ€“ Visual RAG โ€“ Many-shot ICL โ€“ Needle-in-a-haystack โ€“ VL Summarization โ€“ Long-document VQA ๐Ÿ“ Lengths: 8 / 16 / 32 / 64 / 128K ๐Ÿ” Benchmarking both thoroughly & effectively!
3
9
1,647
HKUST NLP retweeted
How far are we from Artificial General Intelligenceโ€”and what might follow as Artificial Superintelligence? IJCAI Closing Panel opened by James Kwok, #IJCAI2025 Programe Chair. share.google/DbB2Ky20bCc3B0Qโ€ฆ #Montreal #AI
1
4
9
1,761
๐ŸŽ‰ Congrats to our students on the new #EMNLP2025 paper acceptances! 1โƒฃ LLM Natural-Formal Hybrid Reasoning arxiv.org/abs/2505.23703 2โƒฃ Text-Instructed Image Editing on Medical Domain arxiv.org/abs/2506.01921 3โƒฃ Knowledge Boundary Aware Multi-Compositional Reasoning arxiv.org/abs/2504.21773 4โƒฃ End-to-End Optimized Multimodal RewardRAG
16
84
6,714
Actually, we implemented this kind of capability in our AdaCtrl paper three months ago by injecting difficult-aware tags (i.e., easy, hard, adaptive) to trigger different reasoning behaviors of LLMs. Paper: arxiv.org/pdf/2505.18822
Updates to ChatGPT: You can now choose between โ€œAutoโ€, โ€œFastโ€, and โ€œThinkingโ€ for GPT-5. Most users will want Auto, but the additional control will be useful for some people. Rate limits are now 3,000 messages/week with GPT-5 Thinking, and then extra capacity on GPT-5 Thinking mini after that limit. Context limit for GPT-5 Thinking is 196k tokens. We may have to update rate limits over time depending on usage. 4o is back in the model picker for all paid users by default. If we ever do deprecate it, we will give plenty of notice. Paid users also now have a โ€œShow additional modelsโ€ toggle in ChatGPT web settings which will add models like o3, 4.1, and GPT-5 Thinking mini. 4.5 is only available to Pro usersโ€”it costs a lot of GPUs. We are working on an update to GPT-5โ€™s personality which should feel warmer than the current personality but not as annoying (to most users) as GPT-4o. However, one learning for us from the past few days is we really just need to get to a world with more per-user customization of model personality.
5
9
1,392
HKUST NLP retweeted
Excited to present our work, MMBoundary, at #ACL2025! Come chat with us at our poster session! ๐Ÿ“ Hall 4/5, Session 12: Poster Session 4 ๐Ÿ—“๏ธ Wednesday, July 30 โฐ 11:00-12:30
3
9
19
1,683
HKUST NLP retweeted
If you are at ACL, please talk to our members of @HKUSTKnowComp and @hkustNLP big family
3
52
2,333
HKUST NLP retweeted
๐Ÿคฏ Multimodal LLMs can be confidently wrong. A single early mistake in perception can lead to a completely incorrect answer. ๐Ÿš€Introducing our work, MMBoundary, a new framework to make MLLMs aware of their own knowledge boundaries! ๐Ÿงต Paper:arxiv.org/abs/2505.23224 #ACL2025
4
12
1,345
Find us at the poster booth โœจWed 7/30 11am Hall 4/5 โœจ
[1/n] "๐˜”๐˜ข๐˜ต๐˜ค๐˜ฉ๐˜ช๐˜ฏ๐˜จ ๐˜ค๐˜ถ๐˜ฆ๐˜ด ๐˜ง๐˜ฐ๐˜ณ ๐˜ช๐˜ฅ๐˜ฆ๐˜ฏ๐˜ต๐˜ช๐˜ค๐˜ข๐˜ญ ๐˜ฐ๐˜ฃ๐˜ซ๐˜ฆ๐˜ค๐˜ต๐˜ด, ๐˜ฅ๐˜ช๐˜ด๐˜ต๐˜ช๐˜ฏ๐˜ค๐˜ต ๐˜ข๐˜ต๐˜ต๐˜ณ๐˜ช๐˜ฃ๐˜ถ๐˜ต๐˜ฆ๐˜ด ๐˜ง๐˜ฐ๐˜ณ ๐˜ถ๐˜ฏ๐˜ช๐˜ฒ๐˜ถ๐˜ฆ ๐˜ฐ๐˜ฏ๐˜ฆ๐˜ด." Such ๐™˜๐™ง๐™ค๐™จ๐™จ-๐™˜๐™ค๐™ฃ๐™ฉ๐™š๐™ญ๐™ฉ ๐™ซ๐™ž๐™จ๐™ช๐™–๐™ก ๐™ง๐™š๐™–๐™จ๐™ค๐™ฃ๐™ž๐™ฃ๐™œ is extremely simple and straightforward for the human cognitive process, but ๐—พ๐˜‚๐—ถ๐˜๐—ฒ ๐—ฐ๐—ต๐—ฎ๐—น๐—น๐—ฒ๐—ป๐—ด๐—ถ๐—ป๐—ด ๐—ณ๐—ผ๐—ฟ ๐—ฐ๐˜‚๐—ฟ๐—ฟ๐—ฒ๐—ป๐˜ ๐—น๐—ฎ๐—ฟ๐—ด๐—ฒ ๐˜ƒ๐—ถ๐˜€๐—ถ๐—ผ๐—ป ๐—น๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ (๐—Ÿ๐—ฉ๐—Ÿ๐— ๐˜€), especially across multiple images and videos โ€ผ๏ธ ๐™’๐™๐™ฎ, and ๐™๐™ค๐™ฌ ๐™˜๐™–๐™ฃ ๐™ฌ๐™š ๐™ž๐™ข๐™ฅ๐™ง๐™ค๐™ซ๐™š upon this major shortcoming in future research? ๐Ÿ” ๐Ÿ”ฅ๐Ÿš€ Check out our latest release, โ€œ๐—ฉ๐—Ÿ๐— ๐Ÿฎ -๐—•๐—ฒ๐—ป๐—ฐ๐—ต: ๐—” ๐—–๐—น๐—ผ๐˜€๐—ฒ๐—ฟ ๐—Ÿ๐—ผ๐—ผ๐—ธ ๐—ฎ๐˜ ๐—›๐—ผ๐˜„ ๐—ช๐—ฒ๐—น๐—น ๐—ฉ๐—Ÿ๐— ๐˜€ ๐—œ๐—บ๐—ฝ๐—น๐—ถ๐—ฐ๐—ถ๐˜๐—น๐˜† ๐—Ÿ๐—ถ๐—ป๐—ธ ๐—˜๐˜…๐—ฝ๐—น๐—ถ๐—ฐ๐—ถ๐˜ ๐— ๐—ฎ๐˜๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐—ฉ๐—ถ๐˜€๐˜‚๐—ฎ๐—น ๐—–๐˜‚๐—ฒ๐˜€โ€! ๐Ÿš€๐Ÿ”ฅ ๐Ÿ”— Project page: vlm2-bench.github.io/ ๐Ÿ“‘ Paper: arxiv.org/abs/2502.12084 ๐Ÿ’ปย Github: github.com/vlm2-bench/VLM2-Bโ€ฆ ๐Ÿค— Huggingface: huggingface.co/datasets/Sterโ€ฆ --- Work led by our amazing team of students Jianshu, Dongyu, and Renjie at The Hong Kong University of Science and Technology. Find us at the poster booth โœจWed 7/30 11am Hall 4/5 โœจ
5
11
895
[1/n] "๐˜”๐˜ข๐˜ต๐˜ค๐˜ฉ๐˜ช๐˜ฏ๐˜จ ๐˜ค๐˜ถ๐˜ฆ๐˜ด ๐˜ง๐˜ฐ๐˜ณ ๐˜ช๐˜ฅ๐˜ฆ๐˜ฏ๐˜ต๐˜ช๐˜ค๐˜ข๐˜ญ ๐˜ฐ๐˜ฃ๐˜ซ๐˜ฆ๐˜ค๐˜ต๐˜ด, ๐˜ฅ๐˜ช๐˜ด๐˜ต๐˜ช๐˜ฏ๐˜ค๐˜ต ๐˜ข๐˜ต๐˜ต๐˜ณ๐˜ช๐˜ฃ๐˜ถ๐˜ต๐˜ฆ๐˜ด ๐˜ง๐˜ฐ๐˜ณ ๐˜ถ๐˜ฏ๐˜ช๐˜ฒ๐˜ถ๐˜ฆ ๐˜ฐ๐˜ฏ๐˜ฆ๐˜ด." Such ๐™˜๐™ง๐™ค๐™จ๐™จ-๐™˜๐™ค๐™ฃ๐™ฉ๐™š๐™ญ๐™ฉ ๐™ซ๐™ž๐™จ๐™ช๐™–๐™ก ๐™ง๐™š๐™–๐™จ๐™ค๐™ฃ๐™ž๐™ฃ๐™œ is extremely simple and straightforward for the human cognitive process, but ๐—พ๐˜‚๐—ถ๐˜๐—ฒ ๐—ฐ๐—ต๐—ฎ๐—น๐—น๐—ฒ๐—ป๐—ด๐—ถ๐—ป๐—ด ๐—ณ๐—ผ๐—ฟ ๐—ฐ๐˜‚๐—ฟ๐—ฟ๐—ฒ๐—ป๐˜ ๐—น๐—ฎ๐—ฟ๐—ด๐—ฒ ๐˜ƒ๐—ถ๐˜€๐—ถ๐—ผ๐—ป ๐—น๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ (๐—Ÿ๐—ฉ๐—Ÿ๐— ๐˜€), especially across multiple images and videos โ€ผ๏ธ ๐™’๐™๐™ฎ, and ๐™๐™ค๐™ฌ ๐™˜๐™–๐™ฃ ๐™ฌ๐™š ๐™ž๐™ข๐™ฅ๐™ง๐™ค๐™ซ๐™š upon this major shortcoming in future research? ๐Ÿ” ๐Ÿ”ฅ๐Ÿš€ Check out our latest release, โ€œ๐—ฉ๐—Ÿ๐— ๐Ÿฎ -๐—•๐—ฒ๐—ป๐—ฐ๐—ต: ๐—” ๐—–๐—น๐—ผ๐˜€๐—ฒ๐—ฟ ๐—Ÿ๐—ผ๐—ผ๐—ธ ๐—ฎ๐˜ ๐—›๐—ผ๐˜„ ๐—ช๐—ฒ๐—น๐—น ๐—ฉ๐—Ÿ๐— ๐˜€ ๐—œ๐—บ๐—ฝ๐—น๐—ถ๐—ฐ๐—ถ๐˜๐—น๐˜† ๐—Ÿ๐—ถ๐—ป๐—ธ ๐—˜๐˜…๐—ฝ๐—น๐—ถ๐—ฐ๐—ถ๐˜ ๐— ๐—ฎ๐˜๐—ฐ๐—ต๐—ถ๐—ป๐—ด ๐—ฉ๐—ถ๐˜€๐˜‚๐—ฎ๐—น ๐—–๐˜‚๐—ฒ๐˜€โ€! ๐Ÿš€๐Ÿ”ฅ ๐Ÿ”— Project page: vlm2-bench.github.io/ ๐Ÿ“‘ Paper: arxiv.org/abs/2502.12084 ๐Ÿ’ปย Github: github.com/vlm2-bench/VLM2-Bโ€ฆ ๐Ÿค— Huggingface: huggingface.co/datasets/Sterโ€ฆ --- Work led by our amazing team of students Jianshu, Dongyu, and Renjie at The Hong Kong University of Science and Technology. Find us at the poster booth โœจWed 7/30 11am Hall 4/5 โœจ
4
1
8
1,414
๐Ÿ“ข Join us for the ACL Mentorship Session @aclmeeting #ACL2025NLP #NLProc โ€ข Session Link: mentorship.aclweb.org/scheduโ€ฆ โ€ข Ask Questions: tinyurl.com/y2v2j462 Mentors: โ€ข @May_F1_ (@hkust) โ€ข @d_aumiller (@cohere) โ€ข @vernadankers (@Mila_Quebec) โ€ข @ziqiao_ma (@UMichCSE) โ€ข @WeijiaShi2 (@uwnlp) โ€ข @ZhijingJin (@UofTCompSci) โ€ข @OanaIgnatRo (@SantaClaraUniv)
19
6
46
15,811
Heading out to #ACL2025 in Vienna with six main/finding papers to present! ๐Ÿ‡ฆ๐Ÿ‡นโœˆ๏ธ๐Ÿคฉ Would love to chat about research on multimodal model reasoning and agent, as well as opportunities in my group @hkustNLP. Please DM if you'd like to meet!
1
7
35
2,740
#COLM2025 Our work has been accepted to COLM 2025๐Ÿ˜Š Looking forward to discussing Scalable Oversight and Synthetic Data with old and new friends in Montrรฉal @COLM_conf ๏ผ
๐Ÿš€ Data to pre-train LLMs on are reaching critical bottleneck. ๐˜ฟ๐™ค๐™š๐™จ ๐™ข๐™ค๐™™๐™š๐™ก-๐™œ๐™š๐™ฃ๐™š๐™ง๐™–๐™ฉ๐™š๐™™ ๐™จ๐™ฎ๐™ฃ๐™ฉ๐™๐™š๐™ฉ๐™ž๐™˜ ๐™™๐™–๐™ฉ๐™– ๐™ฌ๐™ค๐™ง๐™  ๐™จ๐™ž๐™ข๐™ž๐™ก๐™–๐™ง๐™ก๐™ฎ ๐™ฌ๐™š๐™ก๐™ก ๐™›๐™ค๐™ง ๐™จ๐™˜๐™–๐™ก๐™ž๐™ฃ๐™œ ๐™ฅ๐™ง๐™š-๐™ฉ๐™ง๐™–๐™ž๐™ฃ๐™ž๐™ฃ๐™œ ๐™›๐™ช๐™ง๐™ฉ๐™๐™š๐™ง? Let's dive into the "๐—ฆ๐—ฐ๐—ฎ๐—น๐—ถ๐—ป๐—ด ๐—Ÿ๐—ฎ๐˜„๐˜€ ๐—ผ๐—ณ ๐—ฆ๐˜†๐—ป๐˜๐—ต๐—ฒ๐˜๐—ถ๐—ฐ ๐——๐—ฎ๐˜๐—ฎ ๐—ณ๐—ผ๐—ฟ ๐—Ÿ๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€" together! Several key findings: โญ๏ธ ๐—š๐—ฟ๐—ฎ๐—ฝ๐—ต-๐—ฏ๐—ฎ๐˜€๐—ฒ๐—ฑ ๐—ฒ๐˜…๐˜๐—ฟ๐—ฎ๐—ฐ๐˜๐—ถ๐—ผ๐—ป & ๐—ฟ๐—ฒ๐—ฐ๐—ผ๐—บ๐—ฏ๐—ถ๐—ป๐—ฎ๐˜๐—ถ๐—ผ๐—ป of concepts across docs transforms pre-training corpora into ๐˜ฅ๐˜ช๐˜ท๐˜ฆ๐˜ณ๐˜ด๐˜ฆ, ๐˜ฉ๐˜ช๐˜จ๐˜ฉ-๐˜ฒ๐˜ถ๐˜ข๐˜ญ๐˜ช๐˜ต๐˜บ ๐˜ด๐˜บ๐˜ฏ๐˜ต๐˜ฉ๐˜ฆ๐˜ต๐˜ช๐˜ค ๐˜ฅ๐˜ข๐˜ต๐˜ข๐˜ด๐˜ฆ๐˜ต๐˜ด; โญ๏ธ๐—ฃ๐—ฟ๐—ฒ๐˜๐—ฟ๐—ฎ๐—ถ๐—ป๐—ถ๐—ป๐—ด ๐˜„๐—ถ๐˜๐—ต ๐˜€๐˜†๐—ป๐˜๐—ต๐—ฒ๐˜€๐—ถ๐˜‡๐—ฒ๐—ฑ ๐—ฑ๐—ฎ๐˜๐—ฎ reliably follows ๐˜ณ๐˜ฆ๐˜ค๐˜ต๐˜ช๐˜ง๐˜ช๐˜ฆ๐˜ฅ ๐˜ด๐˜ค๐˜ข๐˜ญ๐˜ช๐˜ฏ๐˜จ ๐˜ญ๐˜ข๐˜ธ across various model sizes; โญ๏ธ ๐—Ÿ๐—ฎ๐—ฟ๐—ด๐—ฒ๐—ฟ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ approach optimal performance with ๐™›๐™š๐™ฌ๐™š๐™ง ๐™ฉ๐™ง๐™–๐™ž๐™ฃ๐™ž๐™ฃ๐™œ ๐™ฉ๐™ค๐™ ๐™š๐™ฃ๐™จ; โญ๏ธ ๐˜ฟ๐™ž๐™ข๐™ž๐™ฃ๐™ž๐™จ๐™๐™ž๐™ฃ๐™œ ๐™ง๐™š๐™ฉ๐™ช๐™ง๐™ฃ๐™จ start to kick in near 300B tokens. ๐Ÿ‘‡Check preprint for further details~ ๐Ÿ”— arxiv.org/html/2503.19551v2
6
25
2,293
Excited to share our new survey on the reasoning paradigm shift from "Think with Text" to "Think with Image"! ๐Ÿง ๐Ÿ–ผ๏ธ Our work offers a roadmap for more powerful & aligned AI. ๐Ÿš€ ๐Ÿ“œ Paper: arxiv.org/pdf/2506.23918 โญ GitHub (400+๐ŸŒŸ): github.com/zhaochen0110/Awesโ€ฆ
7
64
5
157
16,168
HKUST NLP retweeted
Thrilled to share a major milestone: the culmination of a 15-month project, ATLASโ€”a new benchmark in event graphs and conceptualization! This journey began with Probase in 2012, evolved through ASER (2019), AbstractATOMIC (2022), and AbsPyramid (2023), and now realizes a decade-long dream of building a web-scale knowledge graph integrating entities, events, and a conceptualization framework. Itโ€™s also a tribute to Marvin Minskyโ€™s K-Lines Theory. I still vividly recall 2012 when I worked with Haixun, including intern Fangting Xia, spent months exploring event semantic primitives and conceptualization, only to pivot due to limitations. Fast forward to the era of large language models, and weโ€™ve finally brought this vision to life! Someone argued that knowledge graphs are obsolete in the age of large models. We disagreeโ€”itโ€™s about having high-quality graphs that scale with these models. Thatโ€™s why we invested HK$1.5M to process Wikipedia, academic abstracts, and 3% of web data from Dolma, creating a 1B-node graphโ€”the best we could achieve. While entities cover 70% of factual queries, adding event semantic primitives boosts coverage to 90%. This is why text decomposition is gaining traction. Our extracted knowledge graph excels on complex factual queries like HotpotQA, MMLU, and FELM, making it the largest open-source GraphRAG dataset available. This work is the result of 10+ students and nearly 10 collaborators pouring their hearts into it. Iโ€™m deeply grateful for their contributions to this meaningful milestone, and I hope theyโ€™ve gained as much from the journey as I have. Big shoutout to anyone curious about pushing this further! Parsing the entire webโ€™s data would cost ~HK$40M, and Iโ€™d love to collaborate with bold thinkers to make it happen. Paper: arxiv.org/abs/2505.23628 Code & Resources: github.com/HKUST-KnowComp/Auโ€ฆ Project: pypi.org/project/atlas-rag/ Feedback and critiques are welcome! Letโ€™s keep advancing the field together. #KnowledgeGraphs #ATLAS #GraphRAG
2
8
2
42
4,476
We studied both rule-based and model-based verifiers and found that each has unique limitations. Rule-based verifiers are often unreliable, even in math, and are unavailable in many domains. Model-based verifiers can be easily hacked. In our paper, we construct simple adversarial patterns to successfully crack several open model-based verifiers. ๐ŸงInterestingly, as the community shifts toward generative verifiers, we found them to be far more vulnerable than discriminative ones. This suggests that discriminative verifiers might be more robust against reward hacking in RLVR.
๐Ÿ” Are Verifiers Trustworthy in RLVR? Our paper, Pitfalls of Rule- and Model-based Verifiers, exposes the critical flaws in reinforcement learning verification for mathematical reasoning. ๐Ÿ”‘ Key findings: 1๏ธโƒฃ Rule-based verifiers miss correct answers, especially when presented in different formats. 2๏ธโƒฃ Model-based verifiers are susceptible to reward hacking, undermining RL outcomes. 3๏ธโƒฃ A probe study reveals that most model-based verifiers, particularly generative ones (like those using chain-of-thought reasoning), are highly vulnerable to adversarial attacks. ๐Ÿ“„Paper: arxiv.org/abs/2505.22203 ๐Ÿ’ปCode & Model & Dataset: github.com/hkust-nlp/RL-Veriโ€ฆ Special thanks to my amazing collaborators @AndrewZeng17, Xingshan Zeng, and Qi Zhu, and to our advisor @junxian_he! ๐Ÿฅฐ
2
17
134
15,709
Great to see the wonderful series of work that @WangCarrey has been leading at UIUC. We also had a fun collaboration recently together with my incoming PhD student Shijue. Check out our latest release ๐˜ˆ๐˜ฅ๐˜ข๐˜Š๐˜ต๐˜ณ๐˜ญ: ๐˜ˆ๐˜ฅ๐˜ข๐˜ฑ๐˜ต๐˜ช๐˜ท๐˜ฆ ๐˜ข๐˜ฏ๐˜ฅ ๐˜Š๐˜ฐ๐˜ฏ๐˜ต๐˜ณ๐˜ฐ๐˜ญ๐˜ญ๐˜ข๐˜ฃ๐˜ญ๐˜ฆ ๐˜™๐˜ฆ๐˜ข๐˜ด๐˜ฐ๐˜ฏ๐˜ช๐˜ฏ๐˜จ ๐˜ท๐˜ช๐˜ข ๐˜‹๐˜ช๐˜ง๐˜ง๐˜ช๐˜ค๐˜ถ๐˜ญ๐˜ต๐˜บ-๐˜ˆ๐˜ธ๐˜ข๐˜ณ๐˜ฆ ๐˜‰๐˜ถ๐˜ฅ๐˜จ๐˜ฆ๐˜ต๐˜ช๐˜ฏ๐˜จ ๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅ . . . arxiv.org/pdf/2505.18822
Time flies. My visiting at UIUC is coming to an end. Iโ€™m sincerely grateful to Prof. Heng Ji @hengjinlp for the invaluable opportunity to join such a warm, collaborative, and world-leading research group. Iโ€™ve learned so much from Prof. Heng and lots of amazing students here @qiancheng1231 @xiusi_chen @emrecanacikgoz @BowenJin13 @zhenhailongW @Yuji_Zhang_NLP @Glaciohound. This experience will always be a defining chapter of my Ph.D. life, filled with friendships, collaborations, and unforgettable moments. Looking back, we truly did something together for the world โ€” from SMART to ToolRL, OTC, RM-R1, ModelingAgent, and DecisionFlow, with even more impactful work on the way. Iโ€™ll always be cheering for BlenderLab and looking forward to whatโ€™s next!
2
9
2
22
8,611
HKUST NLP retweeted
๐Ÿš€ We release MMLongBench: Benchmark for evaluating long-context VLMs. ๐Ÿ“Š 13,331 examples across 5 tasks: โ€“ Visual RAG โ€“ Many-shot ICL โ€“ Needle-in-a-haystack โ€“ VL Summarization โ€“ Long-document VQA ๐Ÿ“ Lengths: 8 / 16 / 32 / 64 / 128K ๐Ÿ” Benchmarking both thoroughly & effectively!
4
32
4
138
23,363
๐Ÿš€ Our new survey "From Automation to Autonomy: LLMs in Scientific Discovery" is live! ๐Ÿ“œ Paper: arxiv.org/pdf/2505.13259 โญ Awesome Github Repo: github.com/HKUST-KnowComp/Awโ€ฆ #LLMs #ScientificDiscovery #AI #NLP #Agent #AIforScience @HKUSTKnowComp @hkustNLP (cont'd)๐Ÿ‘‡
1
5
7
812