Introduction

A few months ago I discovered fast.ai’s fastbook. It was released in August with an accompanying set of YouTube videos that serves as an introductory course in the topic of deep learning. The content is made very approachable by the authors Jeremy Howard, Rachel Thomas, and Sylvain Gugger.

I already had a cursory knowledge of AI having taken a class in college, but I wanted to dive deeper into the inner workings — ideally all the way to the metal. I purchased the book on Amazon in August, but I’ve been slow to go through it since some of the video lectures are rather long, and I learn far better by reading and doing versus listening.

This week I’ve decided to pick up where I left off. I had been thinking about starting a early today (why not), and one of the sections of the book/lecture specifically calls out the advantages of starting your own blog which is rather strange for a book about deep learning. Anyway, that section of the book pushed me over the edge, so now here we are.

To give a quick introduction of myself: My name is Jerred Shepherd. I’m a software engineer working at Amazon Web Services. I work on problems regarding distributed systems and scalability which has been really fun. I love computers and programming, and I often spend my free time working on side projects as a hobby.

Recent posts from blogs that I like

From family to Fantin-Latour: group portraits go weird 1

Before the arrival of photography, they formed an essential family record for children who only saw their mother as infants, royal families, and Dutch guilds. Then they became stranger.

via The Eclectic Light Company

CTO of ‘RansomwareHelp’ Indicted for Wire Fraud

The chief technology officer and co-founder of ransomwarehelp.com, a six year old Florida business that claimed to use "advanced cryptographic tools" to restore data locked by ransomware, has been charged with wire fraud after an FBI sting operation found the company was simply paying the attackers ...

via Krebs on Security

Why output caching doesn't work

LLM inference economics depends on caching1 model inputs, to prevent the entire conversation history being recomputed on each message2. Should we cache model outputs in the same way? As it turns out, no.

Why cache outputs?

First, here’s why we might think caching outputs is a good idea. We cache inp...

via Sean Goedecke