Read time: 6 minutes

Hey guys,

Quick update, we changed the name of the newsletter to The Net Gain! Due to the results of our poll, we decided to change the name. No worries, we will still cover financial content, but we’ll also add on other fun business-related automations!

Now, if you’re looking for a new project to start this weekend, OpenAI just announced that the new GPT feature, which lets you build your own chatbots, is available to everyone.

Here’s what you need to know…

🤖 Tutorial

GPTs 101: Everything You Need to Know

In this guide, you’ll learn everything you need to know about GPTs and their capabilities, but more importantly, gain a clear understanding of the moves you can take right now to position yourself at the forefront of this massive new market

What Are GPTs ❓

OpenAI is calling GPTs “Custom versions of ChatGPT”.

The idea is that you can create a fully customized version of ChatGPT that's perfectly tailored to your own product or company.

They are the first step toward allowing users and companies at large to build their own AI assistants—directly within ChatGPT's interface and via the new Assistants API.

And more importantly, GPTs are the first (official) step towards autonomous agents.

Much of what we can build now may seem basic and not all that useful when humans are required to operate them.

However, this is all a pregame for next year when the GPTs we create now will be operated by other GPTs, potentially revolutionizing the way almost everything is done throughout a business.(more on this later)

How They Work ⚙️

GPTs combine three key ‘ingredients’ to create powerful versions of ChatGPT specific to a use case.

These ingredients are the variables you can control to define what your GPT can do and how it behaves.

Your Ingredients:

• Prompting: You can give it a detailed set of instructions to guide its answers (prompting)

• Knowledge: You can add your own company files for the AI to draw information from.

• Actions: You can use either OpenAI's existing capabilities (like DALL·E, Browse with Bing, or Data Analysis) or your custom capabilities (other actions the AI can perform).

Let’s unpack these a little more…

Behavior 🧑🏽‍🚀

This is nothing new, essentially you can provide written instructions on how you want the GPT to behave. You’ve already been doing this within ChatGPT for months.

Knowledge Base 📚

You’re probably familiar with the concept of a knowledge base and how it allows us to give the model access to external or additional data to use when generating responses.

Up until now, you’ve had to create your own knowledge bases and connect them to a retrieval system with all sorts messing around.

With this feature, OpenAI has build all of this functionality into GPTs natively meaning far less hassle for you and likely much higher quality outputs and overall experience when ‘chatting to data’.

Under the hood, OpenAI are using the same embedding and vector search technology we’ve all been using for months:

This feature is AMAZING news for us as AI Agency owners. Creating custom knowledge chatbots with these rickety third party providers was always a challenge, by using this out of the box we will (likely) have a far easier time allowing our clients to chat to their data in a meaningful way.

Actions

Now this is where it gets really exciting…

Not only can we enable OpenAI’s own powerful features:

• Image generation with DALLE

• Web browsing with Bing

• Data analysis with code interpreter

But we can also equip our GPTs with our own custom capabilities that the AI has permission to execute.

On a basic level, this means our GPTs can be hooked directly into platforms like Zapier and allowed to do things throughout a business.

Yep, all of Zapier’s 6700+ apps can now be interacted with by your GPTs with the right integrations: https://actions.zapier.com/docs/platform/gpt

However, on a more advanced level this allows us to create custom functions that can do ANYTHING across a business via the use of custom functions.

For example, say the user wanted to upload new data to our Airtable. We could connect the following functions and let the AI pick out when to trigger this, instantly processing the input and pushing the new data to Airtable.

def process_data(data):
    # Example processing: capitalize string fields
    processed_data = {key: value.capitalize() if isinstance(value, str) else value for key, value in data.items()}
    return processed_data

def send_data_to_airtable(api_key, base_id, table_name, data):
    # Initialize the Airtable client
    airtable = Airtable(base_id, table_name, api_key)

    # Process the data if needed (this is where you would add your processing logic)
    processed_data = process_data(data)

    # Send the data to Airtable
    try:
        response = airtable.insert(processed_data)
        print(f"Data sent to Airtable: {response}")
    except Exception as e:
        print(f"An error occurred: {e}")

This is a basic example but you get the idea…

You can build some wild stuff and hand the keys over to your GPT to use within conversations as needed.

Jimbo’s Jeans Example 👖

To wrap your head around some of these use cases and how these GPTs can help businesses, imagine a company called “Jimbo’s Jeans”, a jeans manufacturer and retailer. Jimbo’s could utilize the following GPTs:

• Logistics Assistant - A GPT prompted to help staff members to get easy access to logistics data across the business and provide new logistics data to be added to the system. This GPT could be given a custom knowledge of the business’s shipping policies, routes, contacts etc and enabled with custom functions (actions) to be able to not only read and write to the company’s database as needed, but analyze data and provide insights. OpenAI’s data analysis plugin ENABLED.

• Sales Assistant - A GPT prompted to help allow sales reps to log/track their sales activities and get easy access to company SOPs. Provided with custom knowledge of the company’s sales training resources and SOPs and equipped with custom functions that allow the sales rep to interact with the company CRM as they perform their sales activities.

• Financial Assistant - A GPT prompted to help company’s executives better understand their financial data by analyzing trends and identifying outliers while providing handy visualizations of their data as needed. Provided with custom knowledge in the form of many smaller spreadsheets containing aggregate data on key business metrics, equipped with custom functions that allow querying of the company’s database or accounting platform (eg Xero API). OpenAI’s data analysis plugin ENABLED.

• Creative Assistant - A GPT prompted to help the company’s marketing team to create promotional materials that are in sync with their branding and design principles, prompt includes info about who the brand is, who their target market is, what their design principles are etc. Custom knowledge containing examples of recent marketing material that performed well (yes, images) to be used as examples. Custom functions enabled to allow the GPT to schedule new social media posts or modify send created images to the design team’s inbox. OpenAI’s DALLE3 plugin ENABLED for image generation

As you can see, GPTs can be created to assist with specific parts of the business by using different combinations of these ‘ingredients’.

As mentioned earlier, the usefulness of these GPTs when operated by humans is limited. It is only when these capabilities can be integrated into autonomous systems (other GPTs can use these GPTs) that things are significantly more powerful.

Don’t Believe The Hype (yet…) 🤔

While everyone is raving about how crazy all these new features by OpenAI is, its important to try and see through the hype and see things as they really are.

OpenAI have rolled out some pretty ambitious new features which are yet to be battle-tested:

• The new and improved GPT4 128k is not only said to be faster and cheaper but also better able to use the information across larger inputs (eg pickup details in a 200 page book). If you’ve maxed out GPT-4 32K and tried to get get useful outputs you’ll know that it struggles, I’m very skeptical about whether or not a 4x increase would really be able use all of the tokens appropriately and actually be useful.

• GPTs comes with a knowledge retrieval component, a technology which up until now has been plagued with inconsistencies and low quality responses. Getting a perfectly functioning ‘custom knowledge ChatGPT’ has been difficult up until now with the entire world trying to figure it out, will OpenAI really be able to create a knowledge system the does not suffer from their common shortcomings? Especially in the easy-to-use, mass-market-friendly format of the GPTs editor which lacks customizability (right now).

Before you go all in on GPTs, its important to know these limitations have existed for years and while we would love for OpenAI to have found a solution, we should assume that they will encounter the same issues we have but over time be able to solve things with their resources.

From my initial testing of GPTs the Knowledge Base isn’t the best and still is prone to the issues I’ve seen time and time again in building custom chatbots.