The Hive

Share this post

Weekly Update #3

abiraja.substack.com

Weekly Update #3

šŸŽ„ and AI

Abi Raja
Dec 22, 2021
Share this post

Weekly Update #3

abiraja.substack.com
A simple Solana NFT showcase

A quick recap of last week’s goals and progress:

  1. 🟢 Deeper Into Crypto

    a) Wrapped up the Solana NFT not-a-showcase (screenshot above). You can put in anyone’s address and see all their NFTs on the Solana chain. Fun project to learn more about Solana but man the docs to figure out how to do these queries are basically non-existent. Also, everything is a freaking ā€œaccountā€ in Solana. So confusing.

    b) Spent a lot of time following this tutorial to build out an escrow ā€œsmart contractā€/ ā€œprogramā€ on Solana. It’s a wonderfully, well-written tutorial. And I’m really enjoying writing Rust. Just need to find more projects to work on in Rust :)

  2. 🟢 Applied to the OnDeck Founders program!

  3. 🟠 Wrote 1 (Moving Fast), not 2 blogposts.

The two technologies I’m most excited by right now are crypto and AI.

Crypto is a whole lot easier for me to get into. I’ve sorta kept up over the years with the broad state of things.

Re: AI, at Stanford, I did a computer vision class. I also audited CS229, the famed ML class by Andrew Ng. Back then, the state of the art in AI was support vector machines (SVMs), which requires lots of human involvement in determining the right features. AI was pretty primitive and only good at very specific tasks like face detection back then. Deep learning — which really took off after I graduated from college — is the idea of having neural networks with many, many layers. As FranƧois Chollet writes, even though neural networks have been around forever, deep learning couldn’t happen until the late ā€˜00s when a confluence occurred of having tons more data, new neural network techniques and massive speed improvements (for instance, CUDA crossing over from gaming-driven GPU innovations).

Until this week, I didn’t really understand how different deep learning was from SVMs or have any general understanding of what makes current AI techniques so much powerful than the ones in 2012. Sadly, that’s how out of the loop I’m here. I really want to read an entire AI book and do lots of tutorials but for now though, I’m going to resist the urge. I don’t want to spread myself too thin (guilty of such in the past) between crypto and other things such as freelancing. Will have to live with small practical AI experiments for now.

Goals for this week:

  1. šŸ‘©šŸ»ā€šŸ’»Write a post about querying the Solana blockchain (I want to make this as interactive as possible) and also set up a website besides this Substack with stable URLs for content like this that’s potentially useful to lots of people getting into crypto.

  2. šŸ‘¾šŸ‘¾šŸ‘¾ Given it’s Christmas week, this is a stretch goal: I’ve been inspired by this GAN Google Collab notebook. Try it out - it’s fun. Given the prompt ā€œalien friend by Odilon Redonā€, the GAN generates the output image shown below (3 intermediate states shown). Pretty fucking sick. But it takes 2 hours running on Google Collab. For this week, I’d like to generate 5 images by deploying the code to a production environment to understand how fast the generation can be and also, play with how diverse the generated content can be, given different inputs.

Share this post

Weekly Update #3

abiraja.substack.com
Comments
TopNewCommunity

No posts

Ready for more?

Ā© 2023 Abi Raja
Privacy āˆ™ Terms āˆ™ Collection notice
Start WritingGet the app
SubstackĀ is the home for great writing