AI engineers operate at a higher level of abstraction than Machine Learning (ML) engineers or large language model (LLM) engineers and don't necessarily need to know how to build an LLM or a ML model.
AI engineering builds upon ML systems[^1], but with a focus on large scale, ready made models (aka. base models).
The distinct skills that AI engineers need to know include prompt engineering,
A summary of each chapter of "Designing Machine Learning Systems: An Iterative Process for Production-Ready Applications", 1st edition (June, 2022) by Chip Huyen:
From the 2019 essay "The Bitter Lesson" by Rich Sutton:
Summary: AI research shows that leveraging computation through general methods like search and learning is far more effective than incorporating human knowledge. As computational power grows, these methods outperform human-centric approaches in fields like chess, Go, speech recognition, and vision. The key takeaway: focus on scalable computational methods, not mimicking human thought.
I wish every graduate student in AI would read "The Bitter Lesson".[^1]
(WIP)
I tested Gemini as my Pixel Assistant today. I migrated from Google Assistant to Gemini on Android.
I'm fine-tuning large language models. I need to compare fine-tuning Transformer tools. So I asked Gemini. Here's the link to the entire chat below:
Prompt:
You are an large language models expert. You have trained deep learning Transformer models. I want to fine-tune a base model with my domain-specific dataset.
You are facing these challenges:
GPUs Go Brrr by Hazy Research, Stanford, May 2024.
we’re going to talk about what we’ve learned about making GPUs go brr -- and release an embedded DSL, ThunderKittens, that we’ve built to help us write some particularly speedy kernels (which we are also releasing).
small library (DSL?) that we called ThunderKittens that we hope lets us write simple-to-understand clean code that indeed makes gpus go brrr.
I know your timeline is flooded now with word salads of "insane, HER, 10 features you missed, we're so back". Sit down. Chill. Take a deep breath like Mark does in the demo . Let's think step by step:
High-quality data can come from at least 2 sources:
GPT-4o ("o" for "omni") is having the "HER" moment. To put it in another way, Audio[^1] AI models are having their Stable Diffusion moment too.
HER is a 2013 sci-fi movie where Samantha, an AI virtual assistant personified through a female voice. GPT-4o shows us we are ridiculously close to this movie becoming reality.
GPT-4o is OpenAI's new model which can reason across text, audio, and video in real time. It is described as smart, fast, natively multimodal, and a step towards more natural human-computer interaction. It is extremely versatile and fun to play with.
