Chen Gao

I work closely with several leading autonomy and robotics companies in the Bay Area 🚗🤖 and regularly connect with engineers across planning, perception, simulation, and robotics systems. One trend I’m seeing across many teams: strong systems engineers are increasingly in demand. The stack (often C++ / Python) matters, but what matters more is deep domain experience in robotics or autonomous systems, particularly around safety, evaluation, validation, and fault handling. If you’re working in this space or curious about what’s happening across AV and robotics, feel free to reach out. #robotics #autonomoussystems #physicalAI

28

4 Comments

𝗧𝗵𝗲 𝗯𝗲𝘀𝘁 𝗔𝗜 𝗰𝗼𝗱𝗶𝗻𝗴 𝗹𝗲𝘀𝘀𝗼𝗻 𝗰𝗼𝘀𝘁 𝗼𝘂𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿 $𝟲𝟬𝟬 𝗮𝗻𝗱 𝗮 𝗺𝗮𝗿𝗿𝗶𝗮𝗴𝗲 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁. Our VP of Engineering, Xiaofan(James) Luan, was supposed to buy his wife a Dior bag for their anniversary.  Instead, he bought three Claude Code subscriptions and spent the holiday trying to cross-compile 2 million lines of C++. Every fix on one platform broke two others.  $600 later, the only output was "git reset --hard" — and a very cold dinner table.😂 "Make it compile on Windows" is a trap. The real goal was "compile everywhere without hacks" — no AI is going to figure that out for you at 2 am. What worked: constraints before code, review tests not code, bottom-up, one layer at a time. Same task, two days. Then he ran six parallel Claude sessions across three machines with git worktree. The bottleneck stopped being intelligence and started being how fast one person can alt-tab. AI solves exactly the problem you give it. Engineering is in knowing which one to give. His wife is still waiting for that bag. Full story: https://lnkd.in/gtsW_Wvk ——— Follow Milvus, created by Zilliz, for everything related to unstructured data

9

💨 How fast can an autonomous vehicle think? With Alpamayo 1, NVIDIA's 10B-parameter chain-of-thought reasoning model, the distilled version can reason in real time. Hear Marco Pavone, Yan Wang, Yurong You, and Wenhao Ding from our AV Research team break down Alpamayo 1 and what's next for reasoning in autonomous driving. 🔁 Watch the replay: https://bit.ly/4rbog2t

7

1 Comment

Why does an E3 level SWE at Meta make only ~190k/year, while an E8 level engineer makes over ~$2M/year, even though both engineers are ICs and spend the same time at work? I have spent the last 5 years at Meta as an IC. I joined with a little over 15 years of experience, and I’ve worked with many solid engineers in this time. Here is how I think about that compensation jump. 1. Same hours, completely different “unit of work” An E3’s unit of work is usually a task or a ticket. An E8’s unit of work is a multi year problem for the company. E3: “Implement this service, fix this bug, write this feature.” E8: “How do we cut infra cost by 20 percent across this product” or “How do we make this platform safe to scale to 10x users.” One person is paid to execute. The other is paid to decide what is worth executing in the first place. 2. Radius of impact E3 usually impacts a file, a service, maybe a small team. E8 shapes whole orgs and product lines. If an E3 ships something great, the impact is great but local. If an E8 ships the right platform, hundreds of engineers become faster and the company saves or earns millions every year. Comp tracks the area of the circle you influence. 3. Risk and downside protection At junior levels, mistakes are usually contained and reversible. At senior staff levels, a bad call can burn tens of millions or damage the brand. E8s are paid for judgment under ambiguity. They decide which bets the company should not make, which migrations can wait, which “shiny idea” is going to kill reliability. You pay more to people whose good judgment protects you from very expensive failures.  4. They scale themselves This can happen in a few ways. 1. Delegation with ownership They define the shape of the problem, then hand large pieces to other senior and mid level engineers while keeping the bar and direction clear. 2. Knowledge that travels They write RFCs, public comments, FAQs, wikis, internal posts. One answer helps hundreds of people who will face the same issue next quarter. 3. Tools over heroics Instead of unblocking people manually all day, they build tools, libraries or guardrails so others can unblock themselves. One well designed tool can save thousands of engineering hours every year. This is what “scaling yourself” actually looks like. The company pays for that multiplier.  5. Ownership of the “uncomfortable problems” Junior engineers usually work inside a well defined box. E8s take ownership between the boxes. They pick up problems that:  Span many teams and no one really “owns”  Require aligning leaders who disagree  Have product, infra, legal and security angles at the same time Most people avoid those because they are messy, political and slow. Very senior ICs lean into them. That is where a lot of value sits. Continued ↓

201

15 Comments

We (w/ Shansan Gong, Ruixiang ZHANG, Huangjie Zheng, Jiatao Gu, Navdeep Jaitly, Lingpeng Kong) released a family of 7B diffusion language models, DiffuCoder, that specializes on code generation, with a focus on understanding and improving masked diffusion models. A core analysis of DiffuCoder is the autoregressiveness (AR-ness) score, a novel metric that quantifies the causal patterns in decoding, revealing how diffusion models break from strict left-to-right generation for more flexible, non-linear code planning. Recent advances in autoregressive (AR) models dominate code generation, but diffusion-based LLMs (dLLMs) like DiffuCoder offer a promising alternative, especially for complex programming tasks. DiffuCoder explores how these models decode differently—showing less global AR-ness in code tasks compared to math—and how temperature affects both token selection and generation order, unlike traditional AR models. We also introduce coupled-GRPO, a post-training RL method with a coupled-sampling scheme, to reduce performance drops during accelerated decoding, boosting parallelism and efficiency. We use a self-improvement pipeline that leverages AR-ness analysis, coupled-GRPO optimization, and evaluation on benchmarks like AceCode-89k to refine decoding strategies. This approach enables DiffuCoder to navigate diverse code generation pathways and enhance performance with modest computational overhead. Looking ahead, we aim to further leverage Reinforcement Learning to steer code generation through these decoding patterns, with the discrete nature of AR-ness scores providing a foundation for search-based strategies—ideal for the sparse rewards of optimizing complex code structures. Check out our full paper and code for a deeper dive! Paper: https://lnkd.in/gVWU3BDJ Code: https://lnkd.in/gmXTZ_6n Models: https://lnkd.in/gTcKCDr9 #MachineLearning #AI #CodeGeneration #DiffusionModels #NLP

220

5 Comments

just talked to a hardware startup in sf as a team of 2, that just raised $M for their seed round. they've hit: hundreds or pre-orders, millions of impressions on social, and 2nd iteration of product in less than 10 months. ⚡ speed is the most important thing for building. and we're excited to help anyone who wants to test their speed this summer in ai + hardware with the 30 worldwide hack we're hosting with Boost VC, @Flux, Studio 45, and more. we treat building like a sport at worldwide studios, and think there needs to be more open access sprints like this. more info below.

47

1 Comment

🚀 Just wrapped up our ECE241 Digital Systems final project this week! Over the past few weeks, my teammate Jason Guo and I built a fully functional Snake game on the Altera DE1-SoC using Verilog, Quartus, and ModelSim. Key Features: 🕹️ Real-time VGA rendering 🕹️ PS/2 keyboard input with full scan-code decoding 🕹️ Segmented snake with smooth movement 🕹️ Dynamic speed scaling as the snake grows 🕹️ Random food generation using a 16-bit LFSR 🕹️ Collision detection (walls + self) 🕹️ Game-over screen + live score on 7-segment displays This project was a great blend of digital logic fundamentals and hands-on hardware design. We used finite state machines, counters, registers, and more - bringing it all to life through our game! This was an incredibly rewarding project!

52

4 Comments

Got a chance to try the new Zoox by Amazon driverless/autonomous car today. It’s a different experience from Waymo as the car itself is specifically made for autonomous and also it’s still new with a very different form factor. Zoox has no steering wheel and it’s bi-directional, the front and the back are the same. At this time, I prefer Waymo slightly as it feels smoother and more confident; and very easy to call (many cars). The wait times for Zoox were much longer as there’re not as many cars, but overall it’s still a very good experience. Over time, Zoox should increase its “smoothness” and “confidence” through experience and learning. Would love to hear others’ thoughts on Zoox, Waymo, other AVs, etc. In a sense, this is the future of AI, physical AI. PS. Play the video to see the ride. I’ll post photos in comments to see the robotaxi itself. #PhysicalAI, #AV, #AutonomousVehicles, #Zoox #Robotaxi #FutureofTransportation

150

16 Comments

Google & Meta SWE Career Progression, Part 6/9 - Levels 8-10 Welcome to the pinnacle of the IC path, the top 0.5%. As a Principal Engineer (L8/E8), Distinguished Engineer (E9/L9), or Fellow (L10/E10), you’ve become an executive, joining the most influential, visionary, and deeply technical leaders in your company (and industry). >> Principal Engineer (L8/E8) While L7 SWEs solve complex technical problems, E8s solve business problems. Their success is not measured in elegant code, but in company & market impact, revenue generation, cost reduction, risk mitigation. L8s set product-, division-, or company-wide technical direction. Their decisions can impact billions of users, and can be measured in billions of dollars. They spend their days in meetings; reading and writing (documents or posts, rarely code); and mentoring. Working with hundreds of engineers and executives across the company, their primary mode of work is influence. E8s have the necessary context and credibility to ensure engineering teams are building the right things, and prevent them from building the wrong things. >> Distinguished Engineer (E9/L9) If the Principal Engineer's domain is the company–enormous, global scope–a Distinguished Engineer can impact entire industries. They don't just adopt industry standards, they create them. They shape the future, and steer their companies toward it. E9s work on their company's most existential business technical problems. They have enormous breadth and depth, and influence the technical direction of huge parts of the company and industry. They embody their company cultures–moving fast and (sometimes) breaking things at Meta, and showing a healthy disregard for the impossible at Google. L9s create the scope and clarity for senior engineers to do career-defining work. They’re force multipliers, making entire engineering organizations more effective. Given how much company context it takes to work at this level, it’s rare to see L9s whose tenure isn’t over 10 years–or approaching the lifetimes of the companies (27 years for Google, 21 years for Meta). It’s even rarer to see an externally-hired L9; most are home-grown. >> Fellow (L10/E10) There are just a handful of Fellows (and Senior Fellows) at Google & Meta. “Fellow” is a formal recognition from the company that an engineer’s contributions have already fundamentally altered the course of the company or industry. Think Google’s Jeff Dean and Sanjay Ghemawat, who gave the world MapReduce, Bigtable, TensorFlow, Protocol Buffers; or Vint Cerf, the "father of the Internet"; or Meta’s former CTO, Mike “schrep” Schroepfer. Fellows tend to operate outside the official career ladder. They pursue long-term technical vision with little bureaucratic overhead. They define their company’s technical strategy, thinking decade+ ahead. Up next, I’ll switch to the manager track, and cover what it’s like to be a SWE Manager (M1/L6). #SoftwareEngineer #CareerProgression #Google #Meta

55

1 Comment

Everyone’s talking about using LLMs for software engineering… But what about 𝗵𝗮𝗿𝗱𝘄𝗮𝗿𝗲 𝘃𝗲𝗿𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻? Can today’s LLMs generate a full UVM environment for a complex design? 𝗛𝗼𝗻𝗲𝘀𝘁𝗹𝘆 — 𝗡𝗼𝘁 𝗾𝘂𝗶𝘁𝗲 𝘆𝗲𝘁... Will they get there soon? 𝗠𝗮𝘆𝗯𝗲... But here’s the real question: 𝗪𝗵𝗮𝘁 𝗰𝗮𝗻 𝘁𝗵𝗲𝘆 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝗱𝗼 𝘁𝗼𝗱𝗮𝘆? I’ve been experimenting with a side project where LLMs are used for coverage closure. 𝗧𝗵𝗲 𝗶𝗱𝗲𝗮 𝗶𝘀 𝘀𝗶𝗺𝗽𝗹𝗲:  • Feed the model the spec, existing tests, sequences, and coverage report  • Let it identify coverage holes  • Generate new tests/sequences  • Re-run simulations… and repeat A continuous loop until the coverage target is reached. I tried this on my graduation project — and the results were surprisingly strong (Anthropic's models really shined here 😉). So no — LLMs are not ready yet to generate full complex UVM environment from scratch. But they can already act as a powerful co-pilot in one of the most painful parts of the flow. And that’s just the beginning. If you’re interested in the approach or want access to the loop script, feel free to reach out — happy to share and discuss.

173

12 Comments