Bolin Lai

Applied Scientist @Amazon AGI working on LLM and VLM post-training (SFT+RL); ex-MSL; PhD @GeorgiaTech.

[Bolin Lai @bryanislucky]

Our paper was nominated in the Best Paper Finalist of #ECCV2024. I sincerely thank all co-authors. Our work was also reported by Georgia Tech @ICatGT . My advisor @RehgJim will present it on Oct 2 1:30pm at Oral 4B Session, and Oct 2 4:30pm at #240 of Poster Session.@eccvconf

[Georgia Tech School of Interactive Computing @ICatGT]

LEGO can show you how it's done! New @eccvconf work from @bryanislucky, a new generative tool can produce visual images to accompany step-by-step instructions with just a single first-person photo uploaded into the prompt. #wecandothat🐝 @GTResearchNews

https://t.co/SYafClSFng

[James Matthew Rehg @RehgJim]

Excited to be heading to Denver for #CVPR 2026 with 6 papers accepted to the Main conference and 2 for the Findings track.

[James Matthew Rehg @RehgJim]

Very happy to be in Music City for #CVPR2025 My lab is presenting 7 papers, 4 selected as highlights. My amazing students @IrohXu @zixuan_huang @Wenqi_Jia @bryanislucky Xiang Li @fionakryan and postdoc Sangmin Lee are here! @siebelschool @uofigrainger

[Bolin Lai @bryanislucky]

@RehgJim @IrohXu @zixuan_huang @Wenqi_Jia @fionakryan @siebelschool @uofigrainger Thank you! A bit more details for my few-shot image editing paper. Our model can learn a new image editing operation from textual and visual guidance via in-context learning, and apply it to new query images.

Please check our webpage for code and videos: https://t.co/duJP8j3cZ3

[Georgia Tech Computing @gtcomputing]

Howdy from Nashville, ya'll! 🎸🤠

Check out our stars at #CVPR2025, a top @IEEEorg research venue for computer vision experts presenting their work on how computers interpret the world using image and video data!

Tech’s experts will take center stage this week at @CVPR at the Music City Center to share their breakthroughs in computer vision. @GeorgiaTech is in the top 10% of all organizations for first authors and the top 4% for number of papers. More than 2000 organizations have research accepted into the main program.

Tech's first authors include Chengyue Huang, Bolin Lai, Fiona Ryan, Andrew Szot, Lifu Wang, Lex Whalen, and Haoran You. @ICatGT faculty represent the majority of faculty in the papers program. Yeehaw!

Meet all of our experts now 🔗: https://t.co/2m2V2Qatda

#GTComputing #TogetherWeCompute #ChangeTheGame

[Chubby♨️ @kimmonismus]

https://t.co/JtdKjoSofk

[Bolin Lai @bryanislucky]

@LiangJeff95 awesome

[Bolin Lai @bryanislucky]

The full Llama4 will contain 2T parameters. This is quite amazing to learn "billion" is insufficient to describe the scale of LLMs.

[AI at Meta @AIatMeta]

Today is the start of a new era of natively multimodal AI innovation.

Today, we’re introducing the first Llama 4 models: Llama 4 Scout and Llama 4 Maverick — our most advanced models yet and the best in their class for multimodality.

Llama 4 Scout
• 17B-active-parameter model with 16 experts.
• Industry-leading context window of 10M tokens.
• Outperforms Gemma 3, Gemini 2.0 Flash-Lite and Mistral 3.1 across a broad range of widely accepted benchmarks.

Llama 4 Maverick
• 17B-active-parameter model with 128 experts.
• Best-in-class image grounding with the ability to align user prompts with relevant visual concepts and anchor model responses to regions in the image.
• Outperforms GPT-4o and Gemini 2.0 Flash across a broad range of widely accepted benchmarks.
• Achieves comparable results to DeepSeek v3 on reasoning and coding — at half the active parameters.
• Unparalleled performance-to-cost ratio with a chat version scoring ELO of 1417 on LMArena.

These models are our best yet thanks to distillation from Llama 4 Behemoth, our most powerful model yet. Llama 4 Behemoth is still in training and is currently seeing results that outperform GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on STEM-focused benchmarks. We’re excited to share more details about it even while it’s still in flight.

Read more about the first Llama 4 models, including training and benchmarks ➡️ https://t.co/9G3QgVdCkB
Download Llama 4 ➡️ https://t.co/eVomRvEr0w

[Bolin Lai @bryanislucky]

💻The work was done at GenAI Meta. Thank all collaborators at Meta and my advisor @RehgJim for their strong support.🍻 [8/8]

📄Paper: https://t.co/GezWv3GRtO
⌨️Code: https://t.co/8Paxa2axRx
���️Video: https://t.co/cCv6MUeJbr

[Bolin Lai @bryanislucky]

📢#CVPR2025 Introducing InstaManip, a novel multimodal autoregressive model for few-shot image editing.

🎯InstaManip can learn a new image editing operation from textual and visual guidance via in-context learning, and apply it to new query images. [1/8]

https://t.co/duJP8j3KOB

[Bolin Lai @bryanislucky]

🔎In addition, when different exemplar image pairs are used with the same textual instruction, InstaManip can capture the different visual patterns and apply them in editing query images. [7/8]

[AK @_akhaliq]

Alibaba just dropped Wan2.1

open AI Video Generation

#1 on VBench leaderboard, outperforming SOTA open-source & commercial models

Mastery in complex motion dynamics & physics simulation & text rendering

[Max Xu @maxxu05]

My paper RelCon: Relative Contrastive Learning for a Motion Foundation Model for Wearable Data, from my @Apple internship, has been accepted at #ICLR2025! 🎉

We introduce the first IMU foundation model, unlocking generalization across motion tasks. 🏃‍♀️📊

https://t.co/BGVzFjSjxn

[Bolin Lai @bryanislucky]

An awesome gaze model from my labmate @fionakryan!

[Fiona Ryan @fionakryan]

Introducing Gaze-LLE, a new model for gaze target estimation built on top of a frozen visual foundation model!

Gaze-LLE achieves SOTA results on multiple benchmarks while learning minimal parameters, and shows strong generalization

paper: https://t.co/Is2NgrrurO

[Georgia Tech Computing @gtcomputing]

#ECCV2024 has honored this computer vision research as one of 15 Best Paper Award candidates 🎉! Congrats to the team and lead author Bolin Lai, PhD student in Machine Learning at @GeorgiaTech.

[Bolin Lai @bryanislucky]

@LiangJeff95 我是真菜，jeff整不动我应该也不太行🌚