Lin Tan
@Lin0Tan
Followers
655
Following
1K
Media
20
Statuses
180
Elmore New Frontiers Professor @PurdueCS, Amazon Scholar @awscloud | Ex @Meta @UWaterloo @IllinoisCS @MSFTResearch @IBMResearch | #SE #LLM4Code #AI #Security
Joined August 2024
Can #LLMs replace developers? Introducing RepoCod-Lite 🐟 for faster evaluation to answer this: 200 of the toughest #RepoCod #code-generation tasks:.- GPT-4o and other LLMs have < 10% accuracy/pass@1 on RepoCod-Lite tasks.- Leaderboard - 67
Can language models replace developers? RepoCod says “Not Yet”, because GPT-4o and other LLMs have <30% accuracy/pass@1 on real-world method-level code generation tasks. Leaderboard #LLM4code #LLM #CodeGeneration #Security.@cerias @PurdueScience.
3
16
74
RT @iclr_conf: Policies on LLMs Usage at ICLR 2026 have been released: . Policy 1⃣ Any use of an LLM must be disclosed.Policy 2⃣ ICLR auth….
0
39
0
RT @ruqi_zhang: Purdue IPAI is hiring postdocs in AI! If you're interested in statistical machine learning or trustworthy AI, and would lik….
purdue.edu
Purdue connects emerging leaders to world-class experts in physical artificial intelligence and applied fields through the IPAI Postdoctoral Fellows Program. Applications Due: September 1…
0
2
0
RT @tomzimmermann: 🌟 Call for Nominations: IEEE CS TCSE Awards 🌟.Recognize Service, Education, Synergy, WISE Leadership, Rising Star, Lifet….
tc.computer.org
0
2
0
RT @zijianwang30: After three successful runs of #DL4C at ICLR’22 (remote), ICLR’23 (🇷🇼/remote), and ICLR’25 (🇸🇬), I’m thrilled to announce….
0
6
0
RT @tomzimmermann: 🚨 Submission site is now open for the 2nd cycle of @icseconf 2026 in #Brazil! The deadline for complete papers is 🗓️ Jul….
0
9
0
“Purdue named the most recognized public university in U.S., No. 6 overall in Global University Visibility rankings”!.
A big rankings week for @LifeAtPurdue! Extremely proud that we have been named the No. 1⃣ most recognized public university in the nation and climbed to No. 6 overall in the world in the latest Global University Visibility rankings. 🚂.
0
0
5
RT @cindy_rubio: Are you interested in serving on the Program Committee for @issta_conf 2026? Please let us know by filling out this form:….
docs.google.com
Until 30th June 2025, please indicate your interest to serve on the ISSTA 2026 program committee through filling out this form. Please include as much information as possible. After submitting the...
0
6
0
RT @ACM_President: @TheOfficialACM Council has reaffirmed yesterday that all ACM publications and related research artifacts in the ACM Dig….
0
16
0
RT @ruqi_zhang: Excited to share our latest work on self-correcting reasoning in Vision-Language Models! . - Improve reasoning with minimal….
0
5
0
RT @b_roziere: We released Devstral. It is a 24B model released under the Apache 2.0 license. It the best open model on SWE-Bench verified….
huggingface.co
0
16
0
@yiwu5cs @ieee_ras_icra @PurdueCS @anikbera 💡SELP presents three key insights:.1️⃣ Equivalence Voting, .2️⃣ Constrained Decoding, and .3️⃣ Domain-Specific Fine-Tuning, for safe and efficient robot planning. Check out the SELP Demo video:
0
2
3
@LiangShanchao @NanJiang719 @huyiran1007 @aclmeeting @PurdueCS @cerias 4. WAFFLE Preprint, code, data, and models - Details:
github.com
For ACL25 paper "WAFFLE: Multi-Modal Model for Automated Front-End Development" - by Shanchao Liang and Nan Jiang and Shangshu Qian and Lin Tan - lt-asset/Waffle
Wanna generate webpages automatically? Meet🧇WAFFLE: multimodal #LLMs for generating HTML from UI images! It uses structure-aware attention & contrastive learning. #LLM4Code #MLLM #FrontendDev #WebDev @LiangShanchao @NanJiang719 Shangshu Qian
0
0
1
@LiangShanchao @NanJiang719 @huyiran1007 @aclmeeting @PurdueCS @cerias 3. 🧇 WAFFLE: Webpage Multi-Modal Model for Automated Front-End Development. Wanna generate webpages automatically? Meet WAFFLE: hashtag#multimodal hashtag#LLMs for generating HTML from UI images! It uses structure-aware attention & contrastive learning.
0
0
1
@LiangShanchao @NanJiang719 @huyiran1007 @aclmeeting @PurdueCS @cerias 2. We need better approaches for code generation!. - RepoCod Leaderboard, dataset, and preprint: - The top 2 are DeepSeek-V3 and GPT-4o. - Details:
lt-asset.github.io
Real-world, complex, repository level code generation benchmark
Can #LLMs replace developers? Introducing RepoCod-Lite 🐟 for faster evaluation to answer this: 200 of the toughest #RepoCod #code-generation tasks:.- GPT-4o and other LLMs have < 10% accuracy/pass@1 on RepoCod-Lite tasks.- Leaderboard - 67
0
0
3
@LiangShanchao @NanJiang719 @huyiran1007 @aclmeeting @PurdueCS @cerias 1. 🐟 GPT-4.5 Preview, 4o, OpenAI-o1, o3-mini, DeepSeek AI-R1, V3, Anthropic Claude 3.7 Sonnet, and other LLMs have <30% accuracy/pass@1 on real-world code generation tasks. On 200 of the toughest tasks from RepoCod, none of these models reaches more than 10% accuracy/pass@1.
0
0
1