What is ChatGPT, really?

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:

• Digestible explanation of ChatGPT's technical breakdown

• Step by step guide to creating your own sales call follow up tool with AI & automations

• Curated AI tools and news stories to check out

Let’s get to it! 👇

TODAY'S PERSPECTIVE

A non-technical technical breakdown of ChatGPT 🥸

You realize that ChatGPT hitting 1M users in 5 days, a milestone that took Facebook 10 months, is a big deal.

But where did this come from? Why did we spend our holiday break contemplating the meaning of life? How do we decide where to go from here?

Well, an informed decision is a better one. Let’s start with breaking down what ChatGPT technically is (without getting too technical).

btw - here’s that primer on AI terms if you need it.

ChatGPT is built on top of Open AI’s more general model, GPT 🧱

GPT is a massive (think 175 billion parameters massive) AI model that has been trained on a gigantic dataset of text including a large portion of the internet, wikipedia and books.

Every AI is trained on a goal - and GPT's goal is to predict the next word in a series of words, given the words that came before it.

btw - this matters because you can actually use and customize GPT yourself. We’ll cover that in a future email!

ChatGPT’s secret sauce is that it was fine tuned to be good at chatting 💬

A common approach to adapting AI models like GPT to new tasks is to add a custom layer that's been fine tuned on top of the pre-trained model.

Turns out, if you teach GPT-3.5 to be good at chatting, it’s scary good. (Hence viral moment).

But the good news for the rest of us is that the secret sauce isn’t so secret 🤫

Open AI published a technical step by step to how they built ChatGPT on top of GPT-3.5.

And it’s a process that we can all get a little inspiration from 🦄.

Here’s the gist 👇

1. Open AI built a “dialogue dataset” which consisted of example chat conversations that were created by a team of people they called “human AI trainers” and fine tuned GPT-3.5 to be good at chat

2. Human reviews then ranked example outputs to teach the AI what’s good or bad. Those results were used to add another layer on top of GPT-3.5 called a reward model to teach the AI to give better results

3. Last, Open AI has another system that continues improving the model based on positive and negative feedback. That’s what ChatGPT is collecting from it’s public beta today

Ok that was a lot - and how does that help with the whole identity crisis problem? 🙏

First, hear us when we say that ChatGPT is not voodoo magic.

Even if some of what’s above went over your head - I promise if you stick with us for a bit, it won’t.

In fact, the more you can wrap you mind around teaching a computer to do something (like you’d train an intern, or manage an employee), the less you need to worry about “what’s next” with AI.

And we’re here to help.

USE CASE DEEP DIVE

Edit, don't draft, your follow up emails this year ✏️

What if your sales call follow up emails drafted themselves and all you had to do was edit and send?

2023 is the year for using GPT and automation software to completely transform your workflows and give yourself more time to think, strategize, and be creative.

We put together a step by step guide on using automation software, Make, to begin integrating GPT into your repeatable, daily tasks.

How it works?

1. Post your sales call notes in a Slack channel

2. Receive a follow up email draft generated by AI ✨

All you need to get started is a Slack account, an Open AI account, and a free Make subscription.

We understand that everyone takes notes differently and each team has different sales strategies. And while this guide is a great starting off point for automating workflows, most real life implementations will require prompt customization. That being said, be on the lookout for future resources in which we will provide guidance on prompt engineering and customization for your specific use cases.

✨ Here's the step by step guide to automate meeting follow ups! ✨

Note: we are affiliated with Make, but they also have the best free tier + a really easy to use UX!

LINKS

For your reading list 📚

A useful tool + a fun tool ... you decide which is which

• Chatsonic is a variation on ChatGPT that can search the internet, is fully up to date, generate both text and images, work off voice commands, and more. Free trial with a flexible pricing model for your needs. Yeah, you’re welcome.

• Want to spice up that weekly team meeting? This fun new tool uses AI to generate custom zoom backgrounds and is guaranteed to make you a little more popular at work* *ok not really, but maybe! but we had fun with it

(Affiliated with neither)

What’s ahead in 2023

• Microsoft and Open AI are working to integrate ChatGPT-like functionality into Bing, which will certainly be a big move against Google

• A compilation of rumors about what to expect with GPT-4 (a little old, we’re working on an updated one)

And if you're really nerdy ...

• An in depth discussion on how AI can help with loneliness, but isn’t yet ready to tackle something as critical as mental health

• The founder of Gumroad built a chatbot that can do Q&A with customer support docs in 87 minutes… and shared his code!

That's all!

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

Cheers,

🪄 The AI Exchange Team