Activity
3K followers
-
Jiqi Yang reposted thisJiqi Yang reposted thisWe are expanding the Multimodal Machine Intelligence (MMI) team in Apple! At MMI, we are developing foundational multimodal LLMs and exploring the intersection of sensor intelligence and real-world applications. We believe the next frontier of AI lies in how models perceive and interact with the physical world. We are looking for builders and thinkers to join us for the following roles: Research Scientists: Deep dive into multimodal LLM, conversational agents, world models. https://lnkd.in/gd_Htzn9 https://lnkd.in/gNAckrz3 Applied Researchers/Research Engineers/Context Engineers: Bridge the gap between SOTA research and high quality product. https://lnkd.in/gRrXfchK https://lnkd.in/gjEgyKm7 https://lnkd.in/gphfBiuA https://lnkd.in/gpBZifAU Software Engineers: Build the infrastructure and frameworks that bring these models to life. https://lnkd.in/gjU3h_MT https://lnkd.in/gqfQy4Mw https://lnkd.in/gA9ECsJH https://lnkd.in/gZeDxrXq We are looking for interns in the above space as well, please reach out if you are interested.
-
Jiqi Yang shared thisExcited!China's Pony.ai nabs $102M at nearly $1B valuation to take its self driving platform up another gearChina's Pony.ai nabs $102M at nearly $1B valuation to take its self driving platform up another gear
-
Jiqi Yang shared thisNEU SF?🎉Jiqi Yang shared thisGet started with our Data Mining class in San Francisco and earn credits towards a Master’s degree in Computer Science. https://lnkd.in/gv5q6FJ
Jiqi Yang commented on a post
2y
Congrats Jeff! Glad to hear you’re coming back!
-
Jiqi Yang liked thisJiqi Yang liked thisI will be attending CVPR this year with Aurora, our Perception team is looking for talented engineers and researchers who want to apply their skills to solving hard technical problems with real-world impact, I also still have an open intern role on my team for this summer. if either of these are of interest to you come talk with us about careers in self-driving technology at Booth #215.
-
Jiqi Yang liked thisJiqi Yang liked thisCosmos 3 goes live. 🚀 Cosmos 3 is a family of omnimodal world models for Physical AI — a mixture-of-transformers (MoT) architecture that jointly understands and generates language, image, video, audio, and action. It sets a new state of the art across a broad suite of understanding and generation tasks, ranking as the top open-source model for T2I, I2V, robot policy, etc.. It's been about six months since I joined the team, and I feel grateful looking back on it. This was also my first time going through the full journey — from model training all the way to open-source release. There's something quietly special about seeing it through end to end, and doing it with a team this thoughtful and dedicated made all the difference. A heartfelt thank you to everyone who made Cosmos 3 happen. I hope you enjoy what we've built. More to come. 🌐 Website: https://lnkd.in/gi6Up4HR 📄 Paper: https://lnkd.in/gni2Q9Ha 🤗 Model: https://lnkd.in/gwj4ApDy 💻 Code: https://lnkd.in/gqbV3299 #PhysicalAI #OpenSource #AI #Robotics #NVIDIA
-
Jiqi Yang liked thisJiqi Yang liked thisThis Saturday (06/06) in New York City, BeFreed is hosting a Consumer AI TECH WEEK by a16z gathering alongside 645 Ventures, Fish Audio, Creed and many others. Spent nearly 10 years living in New York City, I personally feel New York is one of, if not the only city with an unmatched energy for consumers, an incredible density of world-class talent a constant passion for building, and fierce competition that pushes the absolute best in consumer to the top. Over the past 18 months, AI has driven some of the fastest-growing products in tech history. Startups have gone from zero to millions of users in record time, completely rewriting the playbooks for product, growth, distribution, and retention. The question is no longer if AI will reach consumers, and more about who will build the products people use every single day. This is an exclusive Tech Week event TECH WEEK by a16z hosted specifically for founders and builders in the Consumer AI sector. Sign up now and link in the comments!
-
Jiqi Yang liked thisJiqi Yang liked thisToday we’re announcing HyperLoop, a verification-driven system for mainframe modernization. HyperLoop marks the culmination of everything we’ve been working on since starting Hypercubic. You can’t just throw AI at mainframe modernization and expect to get very far. Mainframes still run core banking, payments, insurance, government systems, airlines, and more. Decades of COBOL, JCL, BMS, and CICS encode the business logic behind the world’s largest institutions. These systems still work. But they are increasingly hard to change, hard to staff, hard to integrate, and hard to explain. That is what makes modernization necessary. But modernization is risky. A translated program can compile and pass tests while still behaving differently from the legacy system under real operating conditions. In critical workloads, one divergent calculation can be catastrophic. HyperLoop combines AI-powered development with layered verification loops to modernize mainframe applications safely. First, HyperLoop extracts business rules from COBOL and turns them into tests. Then it runs the modern application against a verified simulation twin of the legacy system. Finally, it validates against the real mainframe using production scenarios, captured traffic, adversarial tests, and a comparator harness. With HyperLoop, modernization becomes less like a rewrite and more like a sequence of verified changes. More info in the comments!
-
Jiqi Yang liked thisJiqi Yang liked thisI’m reading and deeply resonate with this page about Elon’s engineering “Algorithm”: 1. Make requirements less dumb 2. Delete 3. Simplify 4. Accelerate 5. Automate What struck me most is the order. In many of my previous projects, including ML infrastructure work, engineers (myself included) often started from the opposite direction: - automate first - make it faster - scale it - then eventually realize parts of the system were unnecessary or solving the wrong problem Only after going through all that pain do we finally understand the problem clearly. I see this pattern constantly in ML: • Automating labeling before understanding the actual failure modes • Optimizing throughput before knowing whether the data is even useful • Building sophisticated evaluation pipelines before defining what “good” actually means • Creating complex infrastructure around assumptions that later turn out to be irrelevant The same thing happens in startups: • Automating sales before finding product-market fit • Scaling infrastructure before real usage exists • Building generalized platforms before discovering the core wedge • Over-engineering architectures before identifying the real bottleneck Sometimes the fastest path is: understand more, delete more, simplify more. Only then automate.
-
Jiqi Yang liked thisJiqi Yang liked thisI’m honored that our research work 10 years ago received the Test of Time Award at EuroSys 2026. It’s especially meaningful to see work from years ago recognized by the community. I’m grateful to my co-authors, advisors, and everyone who built on or discussed this work over the years. Thank you to the EuroSys community for this recognition!
-
Jiqi Yang liked thisJiqi Yang liked thisToday we launched OpenClaw for Sales. It combines Openmart proprietary local business data with LinkedIn data and other best-in-class data sources, helping you go from: market question → qualified leads → outreach all in one workflow. Teams like Whatnot, DoorDash, Alibaba, and many others are already using it. DM me for an invite code, or register here: https://lnkd.in/gxUdJryN
-
Jiqi Yang liked thisJiqi Yang liked thisPeople always ask how we closed deals with DoorDash, Whatnot, and Clay when we were nobody. So I’m spilling the tea, let’s go! First, we found warm intros through actual relationships where someone we knew could vouch for us to someone they knew. For Clay, it was a mutual friend who happened to mention our data quality at the right moment. For Whatnot, it was a previous manager who made one intro that mattered more than a hundred cold emails would have. For Doordash, it’s a combination of investor intro and inbound. Warm intros get you the meeting, but they don't close the deal. What closed these deals was that we provided the best service they'd ever seen from a vendor our size. We over-delivered on every pilot. When they asked for something by Friday, we delivered Wednesday. We treated their pilot like it was our only customer because at that stage, it basically was. The third part is the one most startups mess up. We closed pilots within three months. We didn't let them drag into six-month evaluations where momentum dies and champions leave. Three months gave us enough time to prove value but not enough time for internal politics to kill the deal. We pushed for decisions, sent weekly updates showing clear progress, and made it easier to say yes than to delay. This pattern has held across every major deal we've closed. The companies that struggle with enterprise sales usually fail at step two. They get the intro, do okay work, then wonder why the deal stalls out. Doing okay work gets you ghosted, but doing exceptional work gets you references that lead to the next deal.
-
Jiqi Yang liked thisShare work we did for last couple years at Meta. So lucky be part of the team!Jiqi Yang liked thisDiscover the work behind a highly efficient inference pipeline for recommendation systems, built with PyTorch. Our latest blog breaks down the design of a production-grade PyTorch inference system that supports efficiency, reliability, and scale across demanding recommendation workloads. 🔗 https://lnkd.in/giAHQXFK ✍️ Lu Fang, Shiyan Deng, Hongyi Jia, Huamin L., Ilina Mitra, Sheng Qin, Zhengkai Zhang, Zhuoran Zhao (赵卓然), Zinnia Zheng #PyTorch #RecommendationSystems #Inference #MachineLearning #OpenSourceAI
Experience & Education
-
Apple
View Jiqi’s full experience
See their title, tenure and more.
or
Courses
-
AI Engineer Training Course
BitTiger DS502
-
Algorithm Design and Analysis
-
-
Algorithms
CS5800
-
Analog Electronic Technology Foundation
-
-
Application & Technology of Intelligence Control
-
-
Application and Wireless Network
-
-
C Programming
-
-
Calculus I/II/III
-
-
Computational Optimal Control Practice
-
-
Computer Systems
CS5007
-
Convolutional Neural Networks for Visual Recognition
Stanford CS231n
-
Data Visualization on Cross-Media
-
-
DataBase
CS5200
-
Deep Learning
Udacity
-
Deep Learning For Natural Language Processing
Stanford CS224
-
Digital Electric Circuits
-
-
Digital Image Processing
-
-
Discrete and Data Structures
CS5002
-
Electric Circuit
-
-
Integrative Experiment of Automation
-
-
Introduction of Logistics Automation
-
-
Japanese II
-
-
Linear Algebra
-
-
Machine Learning
Coursera
-
Management Information Systems
-
-
Mathematical Modeling & Simulating
-
-
Modern Sensing Technology & Process Measurement System
-
-
Numerical Method
-
-
Object-Oriented Design
CS5004
-
Principle & Application of Embedded Processor
-
-
Principle and Interface Technology of Microprocessors
-
-
Principles of Automatic Control I/II
-
-
Probability and Mathematical Statistics
-
-
Signals and Systems
-
-
Software Engineering
-
-
Stochastic Process
-
-
System Science and Engineering
-
-
University Physics (A)
-
Projects
-
Robots Manufacturing and Programming
-
• Designed, manufactured and programmed a supporting underwater salvage robot equipped with self-designed manipulator, a line-tracking tunnel mini-cruiser, an object recognition supermarket assistant drone, a two-wheeled self-balancing robot
-
Math Modelling Training Program
-
• Finished 5 complete models and essays of, namely, human resources management, scenic spot management, Ebola anticipation and Control, pig farm management, impact analysis of academic network, contributed 60% models and 100% programs
• Designated as Honorable Mention in the ICM 2015 and Meritorious Winner in the CAMCM 2014Other creators -
Languages
-
English
Full professional proficiency
-
Chinese
Native or bilingual proficiency
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content