I'm going to tell you something that upsets a lot of people. I started working late nights for $8/hour back in 2005. I was writing HTML pages and hiding from communists that made it all illegal. One day, I kissed my family goodbye, fled my country, and came to the US. I'm done

I teach hard-core machine learning engineering. AI/ML changed my life forever. There has never been a better time to build a career that will set you apart for the next 20-30 years. I teach a program where I show people how to build Machine Learning systems. My program is not

Deep TDA is a new algorithm that uses self-supervised learning to overcome the limitations of traditional dimensionality reduction algorithms. t-SNE and UMAP have long been the favorites. Deep TDA has many advantages over them. Here is a use case about Intel and semiconductors:

AI will be one of the most crucial skills for the next 20 years. If I were starting today, I'd learn these: • Python • LLMs • Retrieval Augmented Generation (RAG) Here are 40+ free lessons and practical projects on building advanced RAG applications for production: 1/4

Salary will not make you rich. You need to build your own thing. I made a ton of money freelancing. I'll show you how to do the same. I came to the US from a socialist country. I was poor. I didn't own a computer until I was 23, and it was in 2001 when I first used the

In 2024, I'll write about the 50 most important lessons I've learned as a Machine Learning Engineer in 2024. One at a time. And I'll share them for free. Subscribe here if you are interested: underfitted.svpino.com

2 billion people use spreadsheets every month. Their lives are about to change. Here is ChatGPT integrated natively on spreadsheets. It's one of the best applications of AI that will spread in 2024. You now have access to: • Generate content based on other cells • Summarize,

Here is an open-source library that makes integrating AI into an application extremely easy. CopilotKit. Star their repository: github.com/CopilotKit/Cop… The library has two components: The first component is CopilotTextarea • It's a drop-in replacement for any textarea field

Take your data and split it into 10 different subsets: The first one with 10% of the data. The second, with 20%. The third with 30%, and so on. Train a model using each subset. Evaluate it and plot the results. The attached image shows two examples, and the conclusions are

I only found this a few weeks ago: • Open the ChatGPT app on your phone • Click on the little headphones icon • Start talking to the model • Keep the white button pressed while you talk • Unpress the button when you finish talking • Keep the conversation going I have been

Companies aren't our families. We are a line item on a spreadsheet. A helpful tool helping pay for our boss' vacation. I understand everybody can't be an entrepreneur, but what about you? Do you want to build somebody else's future forever, or would you rather try to build

Imagine if airlines interviewed pilots by asking them to recite the flight manual in latin instead of flying a plane. That’s what tech interviews are.

LoRA adapters are revolutionary. They have changed how we fine-tune and serve models. Unfortunately, most companies haven't realized it yet. TL;DR: Companies can now start shipping fast and affordable models that can provide a personalized experience to customers. A company

Tesla uses 8-bit integers to run their models in real time. We call this process quantization. It's how QLoRA lets us fine-tune billions of parameters using consumer hardware. Here is everything you need to know about QLoRA in plain English: Only a few companies can afford to

Here's one of the biggest breakthroughs in LLM fine-tuning: Most haven't realized this yet, but anyone can now fine-tune a large model to personalize results to individual customers. Fine-tuning huge models is no longer exclusive to multi-billion dollar companies, thanks to

People are sick and tired of $9.99 online, beginner-level video courses. One year ago, I started teaching a hard-core Machine Learning class focused on my experience in the field. And it's the best advanced-level machine learning class on the Internet. Here is what makes my

Better data is better than better models. Most AI projects never graduate from the demo stage. This is now painfully obvious with Large Language Model applications. One of the main culprits is low-quality data full of incorrect edge cases. Models work by approximating their

I recorded a step-by-step tutorial on building a RAG application from scratch. It's a 1-hour YouTube video where I show you how to use Langchain, Pinecone, and OpenAI. You'll learn how to build a simple application to answer questions from YouTube videos using an LLM. But the

vLLM has blown every other LLM inference method out of the water. vLLM is an open-source library to serve large language models. It uses a new attention algorithm to deliver up to 24x higher throughput than HuggingFace Transformers without requiring model changes. This is a

I recorded a new YouTube video to teach you how to evaluate a RAG application. And we'll do it step by step. Starting from scratch. 8 out of 10 people I talk to are evaluating their LLM-powered systems manually. This is wild! They try a few samples and deploy the system if the