The hidden cost of AI tools

Welcome to another edition of what we’re determined to make the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today:

• Understanding the hidden variable cost of using AI tools

• Explaining the GPT Playground + 101 guide + downloadable cheatsheet

• Emerging trends for the rest of 2023

Let’s get to it! 👇

TODAY'S PERSPECTIVE

The Hidden Costs of AI Tools: understanding variable pricing

AI features are coming to all your favorite products.

Most are rolling out for free, or included in your existing package.

But it’s very possible that will change in the near future. We’re here to help you understand why, and how to think about budgeting for using AI in your work.

The reality is that generative AI tools have a different cost structure than traditional software.

That’s not to say that it’s particularly expensive. It’s just different.

Most text and image AI products actually use AI models built and supplied by companies like Open AI. Open AI charges by usage, specifically by how many “tokens” are in your request.

And no, we aren’t talking about NFT tokens.

Tokens are chunks of language or words. And typically, 750 words equals about 1000 tokens. You can read more about and test the translation between words and tokens on Open AI’s website.

Tokens are the unit of cost for AI products. Open AI charges $0.02 per 1000 tokens for its most powerful GPT-3 model, Davinci - and even less for the other models.

So how does this play out in costs? Is that a lot, or a little?

Let's look at a few examples. An average email of 200 words would cost just over $0.01 to use AI to analyze and then draft a response. Generating a blog post of 1,500 words – including a prompt that might have an outline – would cost somewhere in the $0.06 range to create. And one of the more expensive use cases - analyzing a 30 minute meeting transcription and generating a summary - might cost $0.10.

Unlike most software, today's AI tools do have a usage cost. And that is important to think about as you use AI across your tools, processes and products.

But of course the cost savings can be huge when you compare them to the time it might take to do the same task manually.

USE CASE DEEP DIVE

OpenAI's GPT Playground Explained

If you're already comfortable using ChatGPT, you’re probably wondering what’s the next step to learning how to apply AI to your business or work. And if not, check out our free prompting email mini-course.

Here’s where the OpenAI Playground comes in - and we’ve put together some free resources to help you learn how to use it!

This is the OpenAI Playground.

The Playground is the tool that will jumpstart your journey to building your own version of ChatGPT; or anything you can think of 🤯.

Instead of the chat-like interface of ChatGPT, you get a huge, blank text box where you can input text, click submit, and it generates a response for you in green.

In the Playground, you’re using GPT-3, the AI model that powers ChatGPT. While ChatGPT has been trained to be super conversational, GPT-3 is more general, flexible and customizable, so it’s easier to get it to follow specific instructions.

GPT-3 has a ton of advanced settings, which we cover in depth in our GPT-3 Playground 101 Guide, which are crucial for having more control over the output you get.

The two biggest advantages of learning to use the Playground are (1) its flexibility compared to ChatGPT and (2) its direct correlation to how GPT-3 will work in other tools.

In addition to the full guide to the Playground we linked above, we’ve also created an OpenAI GPT-3 Playground Cheat Sheet for quick and easy reference. Download the PDF here, and feel free to share this with your team or network!

LINKS

For your reading list 📚

Emerging themes and trends for the rest of 2023...

• Check out this article arguing that in 2023, all Fortune 500 businesses will need to hire a Chief AI Officer

• Here's a Twitter thread highlighting how in a fast changing environment regarding AI and content, authenticity is an emerging theme

And if you're really nerdy...

• This is an in depth study of whether reinforcement learning is the answer to the alignment problem

• New research by Meta has been released which highlights an approach called Token Merging to reduce the latency of existing Vision Transformer (ViT) models without the need for additional training (yup, pretty nerdy)

MEMBERSHIP UPDATE

We're launching tomorrow!

Are you on the list yet? Join the growing community of people using AI in their work and products. 

That's all!

We'll see you again on Thursday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected]. 

Cheers,

🪄 The AI Exchange Team