Yimei (Liz) Chen

Watch our Google I/O '24 recap! Exciting developments in AI! Gemini, our most capable model yet, is now available to developers worldwide. It's multimodal, meaning it understands text, images, and video, and has a massive context window, so it can process huge amounts of information. This is a big step towards making AI truly helpful for everyone. Check out the summary and see what you think. #GoogleIO #AI #Gemini

8

1 Comment

Weekly Series: Week 18 <> Actions speak louder than words Ad Tech is far from having fully capable AI agents. The amount of stakeholders in supply path optimization (SPO) makes it tricky to fill in for each role; even for AI. ~~ the jargon list is endless ~~ But glad to be a first-mover from the supply side, automating publisher ad operations. This week is especially exciting as our 'AI bot' finally earns the title of 'AI agent. 'Like most startups, we're taking baby steps, starting with simple things things such as: Adding a website ✅ Adding ad units ✅ Scanning ads.txt ✅ Checking payments ✅ getting there... Every week, I’m amazed by the eagerness of our small team while collaborating on our AI capabilities—not just the techies, but everyone! I’ve been sharing teasers of the UndrAds AI for a few weeks now. But here’s the real deal. Until next Friday! #programmatic #adopsAI #undradsAI #Adtech #publishers

17

You want to do Machine Learning on large scale Marketing solutions ? Check below 🚀

3

For me, we’re in the “fun part” of AI right now — hunting for use cases. LLMs have unlocked a wave of capability, but we’re still figuring out where they fit best. VC Tom Tunguz made a great analogy earlier this year: if today’s AI were a direct report, you’d micromanage it. It’s got sky-high motivation (always ready to answer) but relatively low skill. I like this lens for exploring AI applications. Yes, there is a ton of (deserved) excitement around capital-A, capital-I tech companies, but I’m more excited about the AI-enabled businesses that we can create today. Where do you see opportunities for an army of high-motivation, low-skill “workers” to unlock new business potential? #AI #Innovation #ProductDevelopment #FutureOfWork

2

LLMs are transforming the recommendation industry as we speak! From what Spotify plays next to the trending posts on your TikTok feed, LLMs are making their way on every stage of a typical modern recommendation system. This includes: 1️⃣ Feature Processing 2️⃣ Candidate Generation 3️⃣ Ranking 4️⃣ User Interaction I've just published an article exploring the 𝗹𝗮𝘁𝗲𝘀𝘁 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗼𝗻 𝗵𝗼𝘄 𝗟𝗟𝗠𝘀 𝗮𝗿𝗲 𝗽𝗼𝘄𝗲𝗿𝗶𝗻𝗴 𝘁𝗵𝗲 𝗿𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 behind the services you love and use. 🔗 Check it out here: https://lnkd.in/gcgiG4H7 Let’s discuss how we can leverage these innovations to transform our industries. Share your thoughts in the comments! As always, happy learning! -- 𝘐𝘧 𝘺𝘰𝘶 𝘭𝘪𝘬𝘦𝘥 𝘵𝘩𝘦 𝘢𝘳𝘵𝘪𝘤𝘭𝘦, 𝘢𝘭𝘴𝘰 𝘤𝘰𝘯𝘴𝘪𝘥𝘦𝘳 𝘴𝘶𝘣𝘴𝘤𝘳𝘪𝘣𝘪𝘯𝘨, 𝑙𝑖𝑛𝑘 𝑏𝑒𝑙𝑜𝑤 𝑚𝑦 𝑝𝑟𝑜𝑓𝑖𝑙𝑒 𝑝𝑖𝑐𝑡𝑢𝑟𝑒 ☝️ #MachineLearning #LLMs #RecommendationSystems

98

20 Comments

https://lnkd.in/gQ6RzjiJ Exciting news! Jianlong Zhong (Jay) will be showcasing "AirAgent" at 12PM PT on Thursday, 9/12, 2024, during the "Airflow Summit 2024" in San Francisco. Airflow depends on continuous scanning files on disk and parsing DAG definitions in Python codes, this design pattern does not go well with the typical Blue/Green deployment mode for microservices (especially, the disruption to the long-running stateful jobs/tasks). To effectively make Airflow deployment "continuously", Jay came up with the design pattern of a deployer agent which takes care of 3 major functions: [1] fetch a specific DAG tarball (usually the latest) commit from github as well as the new Docker image (of runtime Python dependencies) if image is modified [2] point the symbolic link to the new DAG directory on the EFS/NFS [3] restart the worker and scheduler one-by-one accordingly AirAgent is currently deploying updates hundreds of times a day on both staging and production environments. It has been a game changer in terms of Airflow development lifecycle and developer velocity at Coinbase. Missed Jay's captivating talk last year in Toronto? Check it out here: https://lnkd.in/gBC_udTg. #airflow #deployment #airagent #cicd #velocity

52

2 Comments

Last week Asian founders did... 🚀 Ka Ling Wu launched Upsolve AI (YC W24), a platform to build and deliver analytics to your end-users via embeddable dashboards. 🚀 Sahitya S. launched Endeavor, an OS for manufacturers to automate key workflows across sales, inventory, and supply chain. 🚀 Kevin Yang launched Serif.ai, an AI email assistant that writes context-informed emails in your personal voice. 🚀 Vishnu Sridhar launched Matter Intelligence, developing advanced sensors to capture extreme-resolution images of natural and artificial materials from space to surface. 💸 David Shim raised a $50M Series B for Read AI, an AI copilot for wherever you work, making your meetings, emails, and messages more productive with summaries, content discovery, and recommendations. 💸 Kevin H. raised an $800k Pre-seed round for EQ Exchange (EQX), a crypto exchange practicing radical transparency by publicly disclosing all financials in real-time, both on-chain and off-chain. 💸 Pranav Ravella raised an $8.7M Seed round for Sapien, a platform to create autonomous coworkers for finance professionals. 💸 Alex Yeh raised an $82M Series A for GMI Cloud, an AI infrastructure company providing GPUs and a platform to optimize inference and fine-tuning.🎙️ Arvind Jain, co-founder of Glean, talks about Glean's enterprise search product and the future direction of the company with Sequoia Capital 🎙️ Akhil Ramaswamy, wrote a substack article about approaches to defensibility for AI agents Links for the last two in the comments ⬇️

45

6 Comments

🚀 Zetta just released a new feature for automated anomaly detection to tell you when your data isn't looking right. Data quality is one of the biggest challenges data teams face and Zetta is solving this so you can trust your data and make confident decisions. Zetta is an all-in-one AI data platform designed to give you instant data clarity and insights—no data team, infrastructure, or building required. Company wide automated analytics and insights can be setup as quickly as one day with Zetta. Here’s what you get with Zetta: Ready-to-use dashboards that track your key metrics Automated insights that tell you what’s driving your numbers Predictive forecasts to help you plan ahead A conversational AI Analyst so you can explore your data with simple questions A secure, reliable, and scalable data warehouse that’s fully managed for you Let's unlock the full potential of your data, together! https://zettadata.ai/ #AI #Data #Analytics #Businessintelligence #Startups #Growth

6

Hi from the future. Well not to far in the future just 16 hours for those on the west coast of the US. With my role at Coupang, I have been working from Seoul and spent 10+ weeks here so far. Here are a couple tips I have for visiting a new country: 1. Embrace the differences and get out of your comfort zone. Eat some new foods, try a new activity and seek out to learn about a custom you are not familiar with. 2. Take mass transit (or whatever common transportation in the area). Be present and soak in the experience and be ok getting lost. Give your self some extra time to get around. 3. Learn, read and watch. Check out a local museum or historical site, read a book about the culture (I read Pachinko and would highly recommend it) and watch movie\tv from the area. I have been blown away by getting to spend so much time in Korea. It’s been a wonderful experience learning about the culture, enjoying a ton of great food and the nature around Seoul. What tips have you? Also I think these tips also work even if you visiting somewhere not that far away. I have out some really cool things about cities not that far from my house.

110

4 Comments

Want some cool general purpose AI tech that has nothing to do with LLMs? Check out The Forecasting Company They’ve built a “predict any time series” machine.  As an example - A big theme park chain can invest in detailed demand prediction ML, so they can predict how many visitors they will get and optimize staffing, maintenance, food production etc.  A small independent tourist attraction couldn’t, until now. To test the tech, I downloaded total hospital emergency admissions for England, excluded 2024 from the data I sent to the model and asked it to predict 2024 admissions. The result, not bad (see graph).  Significantly better than just using last year’s number as a proxy.  Could I have achieved the same or better with a few hours messing about with the data in excel, yes.  Could I achieve much better by modeling it all properly with a custom ML model and tons of clean custom features, definitely.  But it took 5 minutes, the model doesn’t require retraining (by me), and I got a decent response. My prediction, LTSMs (large time series models) are going to be a big deal, and this is just the first one I’ve seen. AI disclosure: No generative AI was used to create this post.

32

6 Comments

Recently, I came across an article detailing DoorDash's successful utilization of large language models (LLMs), such as GPT-4, to significantly enhance their retail catalog. https://lnkd.in/gdWjb9ZM 🔥 DoorDash effectively utilized LLMs to automatically extract attributes and implemented retrieval-augmented generation (RAG) to expedite annotation. Furthermore, they explored the potential of multimodal LLMs to process text and images together for attribute extraction. 🎞 Drawing from our experience with building ReelRemy, a movie recommendation system, I am confident that we can successfully apply similar techniques employed by DoorDash. 💡 By leveraging LLMs, we can seamlessly extract relevant entities and relationships from unstructured data sources such as movie plots and reviews. In line with DoorDash's approach, we can also utilize LLMs to discern a user's movie preferences based on their viewing history and ratings. 💡 To further bolster our movie recommendation system, integrating multimodal LLMs to extract information from visual assets like movie posters, trailers, and textual data will undoubtedly enhance the user experience. 💡 Additionally, we could create an LLM-based agent that can search the internet, read reviews, watch trailers, and gather relevant information to provide comprehensive and reasoned movie recommendations tailored to each user's preferences. #llm #usecase #gpt4 #genai

9

Old school and probably not popular but I still believe products should be customer-satisfaction focused/led rather than technology-led. Use cases for AI should be byproducts of the product improvement process primarily. The thing though is that founders need to fundraise so they're likely to sprinkle some AI on their decks to even look serious. Not sure that's how the most enduring companies have been built. Of course, if sources of capital become AI-biased, founders have little choice. Happy to keep an open mind and dive in when a use-case that can 10x customer experience for Good Tenants customers and users shows up.

5

There’s a lot of value to be unlocked by teaching AI to reason within the enterprise context and doing the last mile work. The ‘wrapper on top of gpt’equivalent critique for saas was ‘ui layer on top of databases’. We know how that ended. “Application layer AI companies are not just UIs on top of a foundation model. Far from it. They have sophisticated cognitive architectures that typically include multiple foundation models with some sort of routing mechanism on top, vector and/or graph databases for RAG, guardrails to ensure compliance, and application logic that mimics the way a human might think about reasoning through a workflow.” #ai #aiagents

14

In the thick of COVID, I was managing the Rideshare Labs team at Lyft and we hired Julia Yan as a post-doc, replacing Sébastien Martin in that 1-year role. She had a strong conviction about a project where we determined how to navigate tradeoffs related to asking users to walk to a pickup or from a drop-off. It turned out to be a very interesting operations research problem in addition to a fun causal inference problem that I could help with (figuring out users would behave if we asked them to walk different distances than our current policy). Now it's finally published in Management Science! The funny part is that this collaboration was mostly remote due to lockdown -- I've only met Julia in person once (six feet apart in Dolores Park!) and that's more than for Sébastien! They are both now rising star professors (at UBC Sauder School of Business and Northwestern University - Kellogg School of Management respectively) and one of my bucket-list items is to have a research team dinner to celebrate this paper. For me, this is special accomplishment because I have submitted quite a few papers to Management Science in my career and this is the first to have made it through the famously rigorous review process.

231

13 Comments

AI & SaaS: A Tale of Two Futures What if Excel and other SaaS apps don’t vanish but evolve, blending seamlessly with AI? In a recent Interview (https://reurl.cc/eGz06j), Satya Nadella, Microsoft’s CEO, suggests the “agent era” will replace traditional applications. That’s Nadella’s vision: AI as a co-star, enhancing our favorite tools—think Excel with Python instead of Excel replaced by something else. But here’s the twist: 🔸 I see AI’s disruptive power surpassing any need for standalone software. 🔸 Why open an app when a smart agent can directly pull, manipulate, and display data on demand? Nadella believes AI + SaaS = Harmony. He sees Copilot orchestrating tasks, quietly boosting productivity. Developers still matter, refining user experiences to work with AI rather than against it. I foresee a more dramatic shift. If AI can generate everything from sales forecasts to text summaries instantly, maybe the spreadsheets and CRMs we rely on will fade away. Where do you stand? 📌 Collaboration or Disruption? 📌 AI enhancing SaaS or AI replacing it? No matter which path emerges, one thing’s certain: AI is rewriting the rules—and the next chapter of SaaS will be far from business as usual. Your thoughts? Let’s discuss. #AI #SaaS #FutureOfWork #Microsoft 

4

Since the YC deadline to hear back for W25 is coming up next week, here’s the story of how we went from 2 students coding in an MIT dorm room to being full-time YC founders building Delve in SF. For all you founders anxiously waiting to hear back, just keep building — you’ll be successful regardless of the decision. -- In Fall 2023, my co-founder Karun Kaushik and I were studying AI at MIT and spending all of our free time building an AI medical scribe to help doctors. We had vaguely heard about YC from a friend who had just gotten accepted but had no plans of applying whatsoever. However when the YC deadline rolled around, we spontaneously decided to put in an application on the very last day. We spent that whole day in Karun’s dorm room typing up the application and filming all the required videos. We finally hit submit the minute of the deadline! By the next day we had already moved on. We went back to the usual rhythm of coding in the dorm room, calling with doctors, attending conferences, etc. If anything, we were even more motivated now to get more traction and be successful regardless. Later that week, while we were eating in the dorm kitchen, I noticed I got an email saying we had a YC interview scheduled. My heart rate spiked. I jumped up out of my seat and immediately told Karun. I never seriously considered YC, but for the first time it hit me that it might actually be a possibility. When the day of the interview rolled around, we did mock interviews with former YC founders and focused mostly on calming our nerves. More important than the actual content of the interview, we wanted to show our excitement for building and hacking together as co-founders. Unfortunately the interview caught us by surprise. The group partners didn't seem to care as much about our actual AI medical scribe. Instead, they seemed more curious about a side idea we had listed in our application that we wrote up in a day. The idea was about streamlining HIPAA compliance (since we had to go through HIPAA for our AI medical scribe). By this point, we were already seriously considering pivoting into this and had made a lot of progress on the idea. The YC interview was our tipping point Not too long after, we got an ominous email titled “YC.” We were freaked out. I could not sleep the entire night. Then, in the middle of class the next day, I got a phone call from Karun with Surbhi on the other end of the line. My heart rate spiked. I ran outside of the classroom. That turned out to be our acceptance call:) -- For everyone out there still waiting to hear back, just keep grinding. Build as if you’ve been rejected and you might just be pleasantly surprised. I’m also always here to help with mock interviews/give advice! We’ve helped a handful of friends through the YC app process and are always here to support other founders!

190

20 Comments