TL;DR

OpenAI have made AI development accessible to the masses, but in many cases, to achieve your desired functionality, you need to incorporate other tools to get it done. Here is a list of 5 tools that you should consider adding to your AI-development stack

Introduction

AI has been everywhere lately. And for good reason. GPT Models by OpenAI have made AI development accessible to the masses. You no longer need to know how to train and run your own models. All you need to do is call an API endpoint with your prompt. So it’s never been easier to build your own AI-powered applications.

However, while GPT-based models are very powerful, they can’t do it all. At some point you'll likely find that to make your AI application truly stand out and achieve your desired functionality you need the help of other tools and services.

So we put together a list of 5 such tools that you should be aware of as they can really help you level up your AI-powered app development.

Livecycle & Preevy 🤩

By way of introduction, this content is from the team at Livecycle, makers of tools that help developers do their best work. Our latest open source offering is called "Preevy" - a tool that quickly provisions preview environments so that devs can easily share their work and get clear feedback fast.

We love to create content such as this because, like open source code, we believe quality developer content should be open and accessible to everyone who can benefit from it.

So feel free to check out our GitHub repository and give us a star. It really helps us to keep creating content like this. Thank you !! 🙏 💖 https://github.com/livecycle/preevy

And now, with the introductions are out of the way…

GPT-3-Encoder

https://www.npmjs.com/package/gpt-3-encoder

Let’s start with a simple but essential NPM package. If you’ve worked with GPT3 before you might have hit the model’s token limit. That can happen when you input too much text for the model to process. To avoid that, we need to limit the input. Unfortunately, a token doesn’t simply represent one character or one word. There’s more complicated logic behind it. So let’s say we have to limit our input to 4096 tokens. How do we do  this?

Enter the GPT-3-Encoder package. It can encode each string into tokens just like GPT does. Then we can easily count the number of tokens:

import { encode } from 'gpt-3-encoder'

const encodedString = encode('This is our example string')

console.log('Length: ' + encodedString.length)

Pinecone

https://www.pinecone.io/

Pinecone is quickly turning into an essential service of the AI developer stack. It is a hosted Vector Database. Put simply, it offers long-term memory for your GPT models. As discussed earlier, all GPT models have a token limit. So you can’t just feed it with endless information.

But many of the best GPT use-cases require just that: endless information as context. So to achieve this, instead of just passing as input to the model, we can create Embeddings. Embeddings are a mathematical representation of text. Embeddings can be created through the OpenAI API. Then those embeddings that represent our data can be stored in the Pinecone Database. When we now want to provide context to a GPT prompt, we can simply query the Pinecone Database to find the most suitable information from our embeddings. This can then be passed into the GPT model as input.

Whisper

https://openai.com/research/whisper

Whisper isn’t an individual company. It’s yet another product offered by OpenAI. Think of it as a cousin to GPT3 that handles speech recognition. And it’s very powerful doing just that. You can call it through the same API that you already use for GPT. And it’s quite affordable. For 1 Dollar you can transcribe more than 2 hours of audio.

The Whisper model truly shines when combined with other AI tech. For example, you could use it to transcribe interviews into text. Then, using Embeddings in Pinecone you could build a GPT-based Chatbot that answers questions from the interview.

Eleven Labs

https://beta.elevenlabs.io/

What’s the opposite of speech-to-text? You guessed it! Text-to-speech. And that’s what Eleven Labs does best. We all know the artificial sounding computer-generated voices. Eleven Labs is nothing like that. Thanks to AI technology they managed to build voice models that sound very realistic. Most of the time it sounds just like a human. Plus, this works in multiple languages. 

But that’s not even the best thing about it. You can also create custom voice models based on existing people’s voices. So if you want to make your application come to life with real voices, this is for you. Speech can be generated through their API. There is a free tier available and multiple subscription tiers if you need higher usage limits.

Helicone

https://www.helicone.ai/

This is a less popular tool that you might not have heard of before. But in our opinion it’s a must have for AI apps that want to scale. It’s a proxy for your requests to OpenAI. But what is that good for? Turns out, a lot of things. This allows the service to cache your requests which increases speed and cuts down cost. Not only that, due to seeing all of your requests, Helicon can also provide you with detailed logs. That can help you much better understand your usage patterns. They offer 100K free monthly requests and charge $1 for per 10K additional requests.

Wrapping Up

Building AI-powered applications has become more accessible than ever before thanks to OpenAI’s GPT. But as we have shown in this article, there are other great tools and services to enhance your AI-powered apps even further. We encourage you to try all of them out in your own projects.

Hope you found this list helpful! Follow us for more dev-centric content.