🤖 You won't believe how far AI has come

See what I learned last week about some specific AI APIs.

Author

Ajay Prakash
October 29, 2022

Hi all!

First of all, apologies for the radio silence. Been thinking about how I should go about writing this and the frequency. I'm still figuring it out but I have something interesting to share so here it is.

As a reminder, my name is Ajay and you signed up for my personal newsletter sometime in 2021 or 2022.

I've recently been diving into AI and the recent advents in technology.

Having done my thesis on creating an AI Teacher's Assistant in 2017, I thought we had a long way to go. However, we're so much further ahead than I expected and though a lot of people are talking about it - the unenthused responses of my team members made me realise only a few people really know.

You probably have heard about it, with all the talk about self-driving cars and digital assistants. But what are these things, really? Artificial Intelligence (AI) is a process of programming computers for intelligent behaviour. This means creating algorithms that can learn and improve on their own, making decisions based on data. In the early days, AI was limited to simple tasks like playing chess or solving mathematical problems.

However, lately - I've been fascinated with the transition of AI going from 'logical' games to more creative pursuits. It was fine when AI was just messing with chess games, driving cars, and things we don't want to do - but now it's started to encroach on art, music, and other areas of creativity. (It's even helping me write this blog a little faster!)

Art is a fantastic example. With the development of Generative Adversarial Networks (GANs), AI is now able to generate completely new pieces of art, that are indistinguishable from works created by humans.

It's not just art - AI is also being used to create entire songs and albums. In 2017, Sony released an album called “12 tracks”, which was entirely composed by AI. The album is a mix of different genres, including pop, rock, and jazz.

As you can see, AI has come a long way in a short amount of time. And it's only going to keep getting better. So what does the future hold for AI?

Last weekend, I was doing a little deep dive into different APIs and I was particularly fascinated by the stuff on Hugging Face and Stable Diffusion. (The open-source stuff!)

Here are some of the APIs I thought were cool and some interesting applications.

Stable Diffusion

This is an open-source image creation API - which can do a few different things such as

  • Generating 3D models from 2D images: Stable Diffusion can take a 2D image and generate a 3D model from it. This could be used for things like architectural rendering or product design.

  • Artistic style transfer: This is where you take the style of one piece of art and apply it to another. For example, you could take the style of Van Gogh's “Starry Night” and apply it to a photo of your house.

  • Super-resolution: This is a technique for taking low-resolution images and upscaling them to high resolution. This could be used for things like increasing the resolution of photos or videos.

There is actually a way to run this model for free, this video explained it really well and I was able to create some cool images with very little understanding. It does require some code literacy but you don't need to code.

Hugging Face

This is an open-source natural language processing API - which can do things such as:

  • Text generation: Hugging Face can generate text based on a given prompt. For example, you could give it the prompt “The future of AI” and it would generate a piece of text about it.

  • AI.Text classification: Hugging Face can take a piece of text and classify it into different categories. For example, it could take a news article and classify it as “politics”, “sports”, or “entertainment”.

  • Question Answering: This is an API that allows you to feed it information which is your 'context' and then ask it a question based on the context.

You can play with this model for free as well, and you could do it directly on the Hugging Face website. (Below explains one of the hundreds of the Question Answering APIs available)

Go here and see the Hosted Inference API on the right of the screen to play with it. For this model, you'll have to give it context which might be a Wikipedia article and the questions you ask would be from the context.

For example, in the context field, you could put.

My name is Ajay, I live in Sydney and like tennis.

Then you can ask questions such as:

  • What is my name?

  • Where do I live? / Where does Ajay live?

  • What does Ajay like / What do I like?

Play around and let me know what you think.

There are now dozens of startups popping up in the applied AI field. They are basically creating wrappers around these open-source APIs. The AI model and user interface make it incredibly easy for users to use. Maybe some of these companies will inspire you:

If you find any interesting AI companies, send them my way and I'll share an updated list in my next email!

Did you get this email forwarded? Subscribe to me for more emails like this!

Friend of the Newsletter: AI-NC

One of my portfolio companies, AI-NC is looking for engineers or engineering students to help with beta testing. Would love your help here! It's a nifty platform and if you know your way around a CAD file, then you may have some interesting insights. See details below:

We're excited to be launching our manufacturability software from Australian Manufacturing Week 2022 for 3-axis milling. We're looking for engineers designing parts to try out our software and test its manufacturability analysis.

Our software automates design checks by highlighting non-machinable features and key design cost drivers and provides suggestions to quickly and easily amend them, complimenting CAD and PDM software.

Check out our demo below, or try it out yourself by signing up and uploading any .step file or using any of our sample parts. Use the links below:

Thanks again! Till next time,Ajay