MLEPath

DeepSeek: 🔹 Sparse Mixture of Experts 🔹 Multi-head Latent Attention 🔹 Model Distillation What does it really mean for the future of AI? I’m joining Rotational this Friday, Feb 7 at 12PM ET to break it all down. Join us live on Zoom! 🔗

Pushback on my latest video—some claim seniority is all that matters in ML hiring. That’s just wrong. What actually gets you interviews? ➜ Pedigree ➜ Referrals ➜ What you can do These are the three biggest hiring signals in industry. Stop listening to people who have never hired anyone.

Neural networks claim to mimic biological learning, but their training is nothing like how humans learn. Imagine if, at birth, you were dumped with 80% of all human knowledge—completely out of order—and never allowed to learn again. That’s AI today. Maybe we need a new approach?

Too many ML teams use demos to impress rather than progress. A strong demo culture reduces uncertainty, builds trust, and moves toward production-ready AI. Here’s how to fix the broken “demo culture” and use it for real progress: 🔗 https://buff.ly/40Z0HPX

An M2 candidate I’ve been coaching just got a FAANG offer… 3 months after their final interview. Sometimes ghosting isn’t a no—it’s just an extremely slow hiring process.

Kevin Van Horn (30+ years in ML, ex-Adobe Sr. Staff) joins the podcast to break down Bayesian probability, optimization, and what most ML engineers get wrong. Insightful, technical, and real—tune in now! 🎧

Most ML decisions aren’t final—but early in your career, they feel that way. The best engineers don’t chase perfect solutions; they design systems that are reversible. Learning to distinguish one-way vs. two-way doors is the key to moving fast in ML. #MachineLearning #TechCareers #AIEngineering

#Meta warns that is will fire leakers in a leaked memo Now that's meta.

Had a great chat with a Senior Staff ML candidate in office hours yesterday—it reinforced something important about ML System Design interviews at higher levels. You need to save time for a deep dive, not just cover all stages. Here’s how I structure my time to maximize impact. #TechInterviews

Having assisted numerous candidates in clearing Meta’s ML System Design interviews, I’ve observed that questions typically revolve around: ➜ Recommender Systems ➜ Harmful Content Detection ➜ Topic Modeling (NLP) To book your personalized mock interview, visit mockdesignround.com #TechInterviews

Dr. Rebecca Bilbro’s open-source advice? ➜ Use OSS libraries ➜ Find issues ➜ Fix them with PRs That small bit of code you always rewrite? Others do too—contribute it upstream. More wisdom from Rebecca on this week’s podcast: https://buff.ly/40ulLMw #OpenSource #MachineLearning

My first big launch at Adobe? I was ready. Pushed the code, ran tests, went home. Next morning: production was down. My change broke everything. I was horrified. But I took responsibility, fixed it, and earned more trust because of it. Failure isn’t the end—it’s part of the job. #Engineering #Tech

Big tech hiring is brutal—you need to pass #LeetCode, but spamming 5000 problems won’t get you there. ➜ Learn what companies actually test ➜ Focus on the right prep, not random grinding ➜ Part 1 of 4: Breaking down MLE coding interviews Read here: https://buff.ly/4jyu7LZ #DSA #TechCareers

NeurIPS 2024 felt different—less academic, more implementation. LLMs alone aren’t enough. Reinforcement learning, better evaluation methods, and more complex architectures are the real future. We’re finally moving past ‘just scale it up.’ #NeurIPS #AI

LLMs are plateauing. OpenAI is struggling for breakthroughs. The ‘just throw more data at it’ mentality is nearing its end. This is great news for ML. We need diverse techniques, not just one overhyped approach. #AI #MachineLearning

Using AI for interview prep? Do it right: ➜ Use it to refresh, not to learn from scratch ➜ Ask specific questions with context ➜ Always verify you understand why the answer is correct AI should enhance, not replace, your learning. #TechInterviews #MachineLearning

ChatGPT’s bullet-point spacing is weirdly distinct. Once you notice it, spotting AI slop gets a lot easier. Are we all paying more attention to this, or just me? #AI #LLMs

Want to be an ML engineer? ➜ Build an ML project for yourself first—make sure you enjoy it ➜ Learn & fill in gaps with courses (you now know what’s missing) ➜ Build something for others ➜ Apply & prep for interviews Half of this is actively building. That’s how you get ahead. #MachineLearning

MLE job hunt tip: Batch applications. ➡ Apply to roles with similar requirements at once ➡ Prep for specific interview types ➡ Expand filters only if needed Avoid burnout and stay focused. #MLEngineer #JobSearch

Once you hit Senior MLE, what’s next? ➡ Architect: Big-picture systems ➡ Tech Lead: People leadership ➡ Problem Solver: Deep technical challenges ➡ Right Hand: Flexible across roles ➡ Manager: Strategy & teams The real question isn’t promotion—it’s leadership. #MachineLearning

OpenAI "needs" trillions of dollars...

My student just crushed the ML System Design interview and landed a Meta E5 offer: Base: $230k Bonus: 15% Sign-on: $80k RSUs: $875k over 4 years (vesting quarterly with no cliff) Proof that hard work and preparation pay off. Huge congrats! #MachineLearning #TechInterviews

At 19, my first ML project came from wanting to find a picture of my dad. Too many photos, no ML experience—but the project drove my learning. You don’t need to wait. Start building. Solve a problem that matters to you and let it guide your growth. #MachineLearning

📊 Conducted 100+ ML System Design interviews at big tech—here’s what I’ve learned: ➜ Mismanaging time = running out ➜ Skipping stages = missing the big picture Stay clear, follow these stages, and nail your interview. Timing is everything ⏳ #MLSystemDesign #AI #TechInterviews #BigTech

At Twitter, I shadowed an ML interview. The candidate chose C++—they were strong, but boilerplate cost them time and impact. ML engineers: use Python. It’s concise, expressive, and the industry standard. Don’t let your language choice hold you back. #Python #MachineLearning

AGI hype scares me—not because it’s imminent, but because we may have "Boy who cried wolf" problem. My take: https://buff.ly/4ha9Yu9 #AI

Deepseek R1 is proof that censoring LLMs is harder than anyone expected. The secrets it reveals? You don’t want to miss this. Watch the video: [link] #AI #LLMs #TechInsights

🚀 Announcing MLE Path Podcast! 🎙️ Candid conversations with senior ML engineers and hosted by an industry veteran (Adobe, Twitter, Meta). 🎧 Available now on Spotify, Apple Podcasts, and wherever you listen to podcasts! #MLEPath #MachineLearning #AI #TechPodcast

Here is a fun video of Deepseek answering a question about Tiananmen Square. #DeepSeekR1

$500B for AI infrastructure seems huge, but is it sufficient? AI’s rapid evolution means today’s top-tier hardware might be outdated before it’s even used. Should we invest heavily now, or wait until the tech landscape stabilizes? #AI #TechInvestment #FutureProofing

Today’s market panic reflects how poorly #AI is understood. Yes, optimization will reduce GPU demand eventually—but is that the story here? Not so sure yet. Markets can stay irrational longer than you can stay solvent. Stay careful. #AI #Investing

🚨 System ML Design ≠ ML System Design Key difference: ➜ System ML Design focuses on optimizing infrastructure (GPU clusters, distributed training) ➜ ML System Design focuses on building end-user systems (recommender systems like YouTube) 🧠 Preparation matters—know what you’re solving for!

You know how some proponents of #AIsafety say "it's probably fine unless we hook it up to nuclear codes directly"? We are all hosed: futurism.com/trump-admin-...

When new year’s resolutions meet AI… I asked ChatGPT for a good all-around workout with free weights and it was able to come up with something quite intelligent. If every muscle that currently hurts on my body is being strengthened, I will be in great shape in no time.

Is 2025 the year you are going to #learnml ? I've been in tech for 14 years and re-learned #machinelearning at least 4 times, here is a 6 minute video on where I would start: youtu.be/zDgb2KcvJSY

AI that costs 1,000x more than human labor, performs worse, and only works on data it’s seen. Bravo, OpenAI. The o3 model on the ARC dataset is wildly inefficient. Less than 20% gain for 100,000x(!) the cost over Kaggle SOTA. And it’s still worse than a STEM grad. #AI #o3 #OpenAI #machinelearning

Breaking into AI/ML? Start here: 1️⃣ Know your “why.” 2️⃣ Solve a problem for yourself. 3️⃣ Learn the basics. 4️⃣ Build a portfolio project. 5️⃣ Prep for interviews. Take it step by step for real progress. More details here: youtu.be/t7tOGXZjhHM?... #AI #MachineLearning #TechCareers #LearnAI #CareerTips

Ever had a project you just kept working on to get it just perfect? #machinelearning projects require a more iterative approach. Read about how to stay sane as a perfectionist in ML: open.substack.com/pub/mlepath/...

New video alert! Ever wondered how a Staff MLE with 14 YOE at Adobe, Twitter, and Meta tackles an #MLSystemDesign interview in real time? I also added notes from an interviewer’s perspective, gleaned from conducting 100s of interviews. youtu.be/9U48NlbOzCU #TechInterviews #StaffMLE #BigTech

mlepath.substack.com/p/mlepath-an...

How I Onboarded as an Staff Machine Learning Engineer in One Month Instead of Six open.substack.com/pub/mlepath/... #onboarding #TechCareers #mlengineer

Life update: I will be devoting a lot more time helping people get into and advance in machine learning/AI #techcareers If you are interested, or know someone who is, help me spread the word: mlepath.substack.com/p/to-new-beg...

I’ll be at #NeurIPS 2024—my first time back since COVID! 🎉 Looking forward to connecting with new ML friends. Are you going too? Let’s meet up and chat! 🚀

Pro tip: Be familiar with the whiteboarding software you’ll use in the interview. Losing time here can cost you the job. ⏳ #TechInterviews

✈️ Interviewing at a FANG company is like flying to the moon. A small misstep, and you’re thousands of miles off course. Understand the problem before you solve it. #MLSystemDesign

🌍 At big tech, your offer isn’t just about the base salary. Bonuses, stock, sign-ons… make sure you know ALL the components. More here: https://buff.ly/3zuYVee #TechSalary #CareerGrowth

Use flashcards to prep for Leetcode interviews. Match the problem to the algorithm and boost your speed. ⏲️ #Leetcode #InterviewHacks