About
A decade+ of experience in software engineering, now a staff software…
Activity
3K followers
-
David Li ☀️ shared this#Coursera #MachineLearning Lesson Plan 68% done 🎉 Like last chapter, this one was really tough but rewarding. I learned about word embeddings in high-dimensional feature spaces, and how the industry uses very simple unsupervised learning (almost unintuitively simple) to train ML models to understand the relationship between words and contexts. Word2Vec was a particularly tricky algorithm to understand, because it trains two word embeddings instead of one. Unintuitively, we don’t train for semantic representation, but instead we train for the relatedness of words. Magically as a side-effect, words with similar semantic meanings will have similar embedding vectors 🪄 Another interesting algorithm I learned was GloVe (Global Vectors for Word Representation). This algorithm is able to produce word embeddings that capture semantic meaning by computing the ratio of word relations and occurrences. Even though most people would never train their own word embeddings, it’s nonetheless super cool to understand how they are derived. #Gemini might be better than #ChatGPT! I tried using Gemini for the first time - when snippets of the #AI response can be verified on Google, it is highlighted green; conversely, when Google Search returns no results or contradictory results, it is highlighted orange. I find this immensely useful as a new student of this large space. Next week I'll finally learn about the famous #attention mechanism! Have a great week everyone ❤️ https://lnkd.in/gdFbiH3u
-
David Li ☀️ shared thisFood for thought 🧠David Li ☀️ shared thisAmazon Q1 FY24: • Revenue +13% Y/Y to $143B ($0.7B beat). • Operating margin 11% (+7pp Y/Y). • FCF $50B TTM. ☁️ AWS: • Revenue +17% Y/Y to $25.0B. • Operating margin 38% (+14pp Y/Y). Q2 FY24 Guidance: • Revenue ~$144-$149B ($150B expected). If you like this, you'll love our website! Hundreds of visuals every earnings season. Join over 100,00 subscribers. 👇 https://lnkd.in/gANHJBa7
-
David Li ☀️ shared this#Coursera #MachinLearning Lesson Plan 65% done 🎉 This was a really tough chapter, but I come away with a foundational understanding of sequence models like Recurrent Neural Networks (RNNs) and LSTMs 😮💨 These temporal models are a bit trickier to understand, because they build on top of the complexities of a standard neural network. What helped was to draw out the vectors and matrices with various dimensions, so that I can see how the inputs and outputs matched up, especially for the hidden state matrices (see my Excalidraw diagrams) What I found also confusing is that the terminology “layer” has different meanings in a standard NN vs. RNN. In traditional neural networks, a layer refers to a set of neurons, whereas in RNNs, it refers to a set of cells (which could also contain sets of neurons). The key idea here is that LSTMs and GRUs offer ways for the RNN to “remember” or “forget” context from previous words. There are no hyperparameters to govern the “forgetfulness”. These parameters are learned by the RNN, and not configured. But then I learned that there is actually a hyperparameter associated with RNNs and LSTMs - timestep 😂 This parameter affects both training and inference, and is used to configure how far back we train the model, so that it can learn the nuanced patterns from context. Alongside the lectures, I also found vcubingx's video on RNNs to be a helpful high-level overview: https://lnkd.in/gZQKyMMf The material took me an extra 3 weeks to understand but I feel I’m making the right call by leaning into the discomfort of not piecing things together right away. I hope that the extra I’ve spent here will set me up for success next month when I implement my own #GPT from scratch. Also - next chapter should feel much easier in contrast! I’ll be learning about Natural Language Processing, and how words can be represented as high-dimensional vectors. Have a great week everyone ❤️ https://lnkd.in/g7u2XMBJ
-
David Li ☀️ shared this#Coursera #ML Lesson Plan 62% done 🎉 I had a few days off and was able to blaze through the course on #ConvolutionalNeuralNetworks 🔥 I really liked how Andrew Ng taught this course. He started with the basics on #ComputerVision and how pixel kernels with certain patterns can detect edges. He then moved on to simple #ConvNets, explaining how as the network gets deeper, the resolution decreases, but the depth increases. This transitioned into an easy understanding of object detection and face recognition. I especially liked how Andrew paid homage to the landmark CNNs of the past and explained how they worked 🧠 Alongside the lectures, I also found 3Blue1Brown’s video on convolutions interesting and helpful to understand kernels: https://lnkd.in/gCimKPAr Now that I’ve finished the 4th course, I can get started on the final course of the specialization! Sequence Models, Recurrent Neural Networks and Transformers 🥳 Have a great week everyone ❤️ https://lnkd.in/gJJWJ8pW
-
David Li ☀️ shared this#Coursera #ML Lesson Plan 50% done 🎉 Since I started my ML journey almost a year ago, I am halfway done 🥹 This is a big milestone - according to Coursera, I watched: - 100+ hours of lectures - 50+ hours of quizzes and labs in Jupyter I feel a deep growing sense of confidence in my understanding of #DeepLearning, and I’m proud to say that I have a foundational, but holistic understanding of supervised learning, neural nets, decision trees, and also some practical experience implementing and deploying ML models 🧠 Today I finished the 3rd course (3/5) in the Deep Learning Specialization, and earned a certificate in Structuring ML Projects & Error Analysis ⚙️ Although less implementational compare to this previous courses, this one offered a lot of practical insights. I learned how to assess the effectiveness of an ML model, and how to optimize its success metrics depending on underfit/overfit, and more importantly - how to train a model when specialized training data is scarce, by leveraging general data using mismatched train/validation distributions, and other techniques like fine-tuning and transfer learning. I’m excited to keep the ball rolling - the next course is Computer Vision and Convolutional Neural Nets! Once that’s finished, I can get started on the final course, which I’m the most excited about: Recurrent neural networks and Transformers. Have a great week everyone ❤️ https://lnkd.in/g8s7429p
-
David Li ☀️ shared this#Coursera #ML Lesson Plan 47% done 🎉 I finished the second course (2/5) in the #DeepLearning Specialization, and earned a certificate in Hyperparameter Tuning, Regularization and Optimization ⚙️ After a hectic Q4 and a successful Q1 product launch at work, I’m happy to resume my regular schedule of studying ML. Coming back after a hiatus, I felt rusty, but still remembered the core concepts. Luckily, given my foundation, the various ways to optimize an ML model discussed in this course felt intuitive, for example, batch normalization as an extension of input normalization, and Adam as an extension of Gradient Descent 📈 Looking forward to playing with an ML flight simulator next week for the course on Structuring Machine Learning Projects. Have a great weekend y’all ❤️
-
David Li ☀️ shared thisHi friends, As you might've heard, #Etsy laid off around 11% of its employees this morning. I was not affected, but I personally know many amazing colleagues who were. If you are hiring, please let me know. I would love to connect you with someone who's kind, talented, and a great fit for the role. (I've linked the original post from my friend, colleague, and mentor Emjek)David Li ☀️ shared thisThis morning, Etsy laid off "approximately" 225 staff. I was not laid off, but I know many great people who were, across a wide range of functions. If you're hiring now or soon, let me know; I bet I know someone who's kind, talented, and a great fit for that role.
-
David Li ☀️ reacted on thisDavid Li ☀️ reacted on thisPlease help me find a job - I’ve been unemployed for too long (2.5 years). Applying to postings isn’t working and I’m not breaking this job market. Long-time recruiters I know say they have never seen a job market this bad in 30 years, including in the worst recessions. In my case it is in fact urgent because I’ve spent through my savings, got evicted and am living in my car. After posting this for months, I still don’t have a job but many people have said encouraging things and I have raised some money via a GoFundMe to survive in the near future (thanks to those people) but I also know it won’t last so I have to keep it going. (link at the end). Here’s the rest of my situation. I was laid off from my job at Google almost 2.5 years ago when there were a lot of layoffs. I keep ATS confirmation emails from job applications in a folder and at two years since the layoff, I had applied to 3700 jobs and emailed 2200 founders. I won’t do a new count. Bottom line, I'm getting nowhere. I DESPERATELY need a regular job (or monthly contract). It probably has to be a REFERRAL from you because of the hiring environment today. Also, FYI, my present job is just the very rare short term contract. Very rare. My background spans high-growth startups and tech giants, with a proven track record of placing specialized talent. I've placed people with all sorts of backgrounds including engineers, finance execs, marketing, artists and others. At Google I focused on security engineers from the FBI, CIA, etc. PLEASE mention me to your friends or hire me. Please. I’m going keep the GoFundMe going because after 2.5 years, I don’t know when this job market will loosen up. I would have preferred a job months ago but finding a job is not happening. https://lnkd.in/geY_6DPi (I will post this daily)
-
David Li ☀️ reacted on thisDavid Li ☀️ reacted on thisFinished the Certificate Program in Applied Generative AI at Johns Hopkins University this month. The curriculum offered a practical, systematic breakdown of current AI architectures. Looking forward to applying these structural concepts to my ongoing work!Certificate Program in Applied Generative AI completion certificate for ABHISHEK CHAUHANCertificate Program in Applied Generative AI completion certificate for ABHISHEK CHAUHAN
-
David Li ☀️ liked thisDavid Li ☀️ liked thisChatGPT just hit $100 million in annualized ad revenue. It took six weeks. Six hundred advertisers. Ads at the bottom of every answer. "Clearly labeled." Not influencing responses. That's what OpenAI says. I keep thinking about this as someone building products that plug into these models. The tool I use to research, draft, and make product decisions now has a second customer. And that customer is paying to be in front of me at the exact moment I'm making those decisions. OpenAI says ads won't influence responses. I believe them today. But $100 million in six weeks creates its own gravity. Ads went from "we're exploring this" to fastest-growing revenue line before most people noticed. This is the pattern every platform follows. Search was neutral once. Social feeds were chronological once. Recommendations were organic once. Then the ad business got big enough to matter. And the product quietly reorganized around it. If you're building on top of ChatGPT or integrating its recommendations into your workflow, the ground just shifted. The AI is still good. But it's no longer just yours. The best time to think about platform dependency was before the platform found its business model.
-
David Li ☀️ liked thisDavid Li ☀️ liked thisCompanies aren't just laying people off due to AI. They're removing entire levels. Your next promotion might not work the way you think. There's a structural shift happening inside major tech companies right now that isn't getting enough attention. It's not just layoffs. Companies are actively removing layers of management and compressing their org charts, and it's changing how careers in tech actually work. Amazon kicked this off in late 2025 when SVP of HR Beth Galetti outlined a plan to "reduce layers, increase ownership, and remove bureaucracy." They've since accounted for over half of all tech layoffs in 2026. But the key detail is that these cuts are structural, not performance-based. They're eliminating positions that won't be backfilled. Meta took it further. Their new applied AI division runs at a 50-to-1 employee-to-manager ratio, double what's traditionally considered the outer limit (25:1). To operate at that ratio, you have to collapse multiple management tiers into one. And this isn't just a Big Tech experiment. Gallup found the average number of direct reports per manager jumped from 10.9 to 12.1 in a single year (2024-2025). Analysts project that by end of 2026, 1 in 5 companies will use AI to significantly reduce their middle management ranks. So what does this mean for people building careers in tech? When companies compress from 7-8 levels to 5, the leveling structure shifts. This would likely mean wider bands, more competition within each level, and fewer stepping-stone roles. And likely generally less hiring overall. This is one of the most significant shifts in tech org design in years. We’re still collecting details on specifics, but we can be sure of one thing: the org charts of 2027 are going to look very different from 2024. Something worth paying attention to as the layoff news continues to roll in.
-
David Li ☀️ liked thisDavid Li ☀️ liked thisI am in shock—my new book is officially a NYT BESTSELLER!!! 🎉🎉🎉
Experience & Education
-
Etsy
View David’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Projects
-
CPSC 319: Pitch'n Volunteer Management System
-
Implementation lead for a volunteer management system in PHP.
Prioritized and assigned tasks to each team member.
Delivered the system's core features to the client on schedule.Other creators -
Languages
-
English
Native or bilingual proficiency
-
Mandarin
Professional working proficiency
Recommendations received
-
LinkedIn User
2 people have recommended David
Join now to viewOther similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More