Jiacheng Zhu
@JiachengZhu_ML
Followers
1K
Following
2K
Media
5
Statuses
136
Research Scientist at @AIatMeta GenAI, Postdoc at @MIT_CSAIL and PhD from @CarnegieMellon | Prev. @Apple AI/ML @MITIBMLab | Working on Llama post-training
Joined October 2019
RT @RickardGabriels: š” Our new work at @ICML2025:.Serve thousands of LoRA adapters with minimal overheadāa step towards enabling LLM personā¦.
0
7
0
RT @balesni: A simple AGI safety technique: AIās thoughts are in plain English, just read them. We know it works, with OK (not perfect) traā¦.
0
108
0
RT @karpathy: Scaling up RL is all the rage right now, I had a chat with a friend about it yesterday. I'm fairly certain RL will continue tā¦.
0
858
0
RT @Hongyu_Lii: We interact with dogs through touch -- a simple pat can communicate trust or instruction. Shouldn't interacting with robotā¦.
0
26
0
RT @_vztu: š¤ØEver dream of a tool that can magically restore and upscale any (low-res) photo to crystal-clear 4K? . š„Introducing "4KAgent: Aā¦.
0
43
0
RT @Happylemon56775: Excited to share what I worked on during my time at Meta. - We introduce a Triton-accelerated Transformer with *2-simā¦.
0
96
0
RT @hardmaru: Reinforcement Learning Teachers of Test Time Scaling. In this new paper, we introduce a new way to teach LLMs how to reason bā¦.
0
123
0
RT @YiqingLiang2: Heading to Nashville for @CVPR ! šø.Iāll be presenting the @nvidia internship project ā. āZero-Shot Monocular Scene Flow Eā¦.
0
6
0
This is an amazing joint-force with @_Jason_Q , @wenhaoding95 @LiuZuxin @jtompkin @mengdixu_ @xiamengzhou @_vztu @ShiLaixi , lead by @YiqingLiang2 !.
š New preprint: āMoDoMoDo ā Multi-Domain Data Mixtures for Multimodal LLM RLā is live!. š Paper + project: (project page . š» Code: coming soon. Highlights. * Multimodal RLVR framework that post-trains an MLLM on 5 vision-language.
0
3
14
š New preprint: āMoDoMoDo ā Multi-Domain Data Mixtures for Multimodal LLM RLā is live!. š Paper + project: (project page . š» Code: coming soon. Highlights. * Multimodal RLVR framework that post-trains an MLLM on 5 vision-language.
lnkd.in
This link will take you to a page thatās not on LinkedIn
š§µāÆMoDoMoDo: smarter data mixing ā stronger reasoning for multimodal LLMs.š New preprint! We show how right mixing of multi-domain data can superācharge multimodal LLM RL reasoning. š Site: š Paper: details in thread. š
1
3
15
RT @YiqingLiang2: š¬āÆI am super excited to announce our @ICCVConferenceāÆ2025 Workshop: EndātoāEndāÆ3D Learning (E2E3D) š¬.& š¢ Call for Papersā¦.
0
10
0
RT @ICL_at_CMU: š Congrats to Prof Changliu Liu @ChangliuL on IEEE RAS Early Academic Career Award in Robotics and Automation "for significā¦.
0
4
0
RT @angelaqdai: What an amazing start to our #Eurographics2025 keynotes from @JustinMSolomon on optimization for geometry processing - lotsā¦.
0
10
0
RT @ShiLaixi: šWe have two exciting works at ICLR!.1) Oral: Behavioral Economics in Decision Making within Robust Multi-Agent Systems ā Comā¦.
0
13
0
RT @ShunyuYao12: I finally wrote another blogpost: AI just keeps getting better over time, but NOW is a special moā¦.
0
188
0
Our first batch of Llama 4 models are out! š¦š¦š¦š¦Please give them a try! š. Proud to be part of team that baked these amazing models! . P.S. LLM post-training is a magic yet not magic. Excited about future Llama models!.
Today is the start of a new era of natively multimodal AI innovation. Today, weāre introducing the first Llama 4 models: Llama 4 Scout and Llama 4 Maverick ā our most advanced models yet and the best in their class for multimodality. Llama 4 Scout.ā¢Ā 17B-active-parameter model
1
0
26
RT @AIatMeta: Today is the start of a new era of natively multimodal AI innovation. Today, weāre introducing the first Llama 4 models: Llaā¦.
0
2K
0
RT @tw_killian: Iām talking to my church youth group this week about how to discern what and what not to trust on the internet + genAI contā¦.
0
2
0
RT @StockMKTNewz: Apple and Meta are both now reportedly planning to ramp up development on Humanoid Robots joining Magnificent 7 names likā¦.
0
458
0