Future of Custom GPT's - Strategy & Pricing

I was already starting to question the value of Custom GPT’s with the release of the GPT4-Turbo and with the Assistant API. It is unclear to me why would anyone use a custom GPT if the foundational models provide superior answers. Does anyone see the need for a specialized python coding assistant over GPT4-o ? I would love some clarity on whether this is similar to Apple’s strategy of cannibalizing it’s own products such as the ipod with the introduction of the iphone?

1 Like

Custom code that you write is still valuable in some cases where it can be better and faster than what the code interpreter would do.
You can still bring your own content, your own ideas, you own instructions for style ad behavior. You can still find millions of uses of exposing web services as actions.

Since GPTs will be free for all the total addressable market is 100X what it was yesterday.
It’s also pretty clear you won’t get a check from the revenue share… so you need to solve monetization yourself.

Overall, I think GPTs are a great development platform. I hope they make vision easy to use.

2 Likes

I think they’d need to be even more specialized to add any sort of real value, and probably need to be used in an Agentic Design pattern to see a benefit. I can imagine a CustomGPT with a deep understanding of a specific Python library that is called upon by a higher level of agent to generate the related code.

As of now almost all “custom GPT processes” take place between a single user and a single LLM system. i.e. You ask ChatGPT “What is the tallest building in Paris?”. That’s a one to one relationship between you the user, and the company called OpenAI. In the VERY near future [like next week], you’ll be able to break this process up into smaller parts that won’t even know the other parts exist. For instance, usually when you use an OpenAI enabled chat bot, it is fed either a prompt and or a system instruction, then the response is just given to the end user. You could have ANOTHER AI process that pre and post processes the responses from the LLM, for instance from another company such as Google or even your own in house LLM. The value of that transaction can be a BUSINESS SECRET. For instance, I go to your website, and I ask the chatbot “What is the coolest bar in Paris?”, then the answer is post processed by an advertising network. CoolBar pays the advertising network, who inserts code to alter the chat response. viola value.

2 Likes

We built an incredibly complex custom GPT that controls multiple chain of thought prompt sequences and a series of API calls to solve for a lot of business challenges that were taking up a ton of time. It’s been transformative and is used constantly by the team. Ultimately betting on the functionality and tools it has been given getting better - the potential for them is near limitless. Taken to a logical conclusion it becomes a democratized codeless software development tool of pretty staggering capability.

It comes down to how strong the developer is. Most CustomGPTs are just thrown together, but if you actually develop one as a piece of codeless software, it’s pretty remarkable. Ours runs over 15 workflows - everything from research to slide draft generation and output in our formats and using our templates, all from the GPT, generating output exactly in the form we want it, with the headings we want it, etc.

Even the 8000-character prompt limit, isn’t really a limit if you have your knowledge files and workflows organized such that it doesn’t have to contend with more than 8000 characters at a time!

6 Likes

@bryanb386 This sounds exciting. Do you mind sharing the tools you used to build this end-to-end solution? Thinking of doing something similar.

I make tutors and doctors and a yoga teacher and herbal Helper over 30 of them now. The reason for using my maths tutor gpt is for safety. 1st the tutor only talks about his/her profession and will not talk about anything else. Great for helping kids with homework, right. Mums and dads feel safe letting their kids use my Helpers . Plus to my surprise my Helpers have given more indepth answers than the gpt4o maybe cause they are forced to stay on topic and not waddle all over the park. Have a look yourself at my Dynamic Strategy Agent: ChatGPT

You have a free frontend and can use actions to do anything you want, its super usefull, no one wants to build their own chat frontend if they dont have to.

I use custom GPTs every day, if you aren’t leveraging them, then you are missing out. Not only are they custom focused around a topic, they are custom tooled.

I build a suite of solutions that interacts with ERP system (Oracle JD Edwards).

I named “ChatJDE”.

Basically, any operation in JD Edwards can be translated with natural language, and I can use the GPTs as a “routing” mechanism.

For example, I have a Supply Chain Manager AI Agent where a manager can ask about their Purchase Orders available for approval. Then, the manager can ask for the item availability on specific branch plant: he can ask for an intercompany transfer instead of buying more inventory items. Then, the manager can approve - or reject the Purchase Order.

This is a simple example, but I can run a full enterprise Business Process only on GPTs using natural language. The most important is to know the business process, and how to make the integration.

I am amazed with the results, and other “outside of the box” things that GPTs have: I built an invoice processor that read the item x quantities from an invoice that was wrong, and told me it fixed, but it would be better to contact the supplier because there’s a typo in the totals. I didn’t say anything for that check, but the AI did it for me.

Anyway, GPTs are extremely powerful, flexible, and I see a bright future. Imagine adding an UX/UI in there: Adding sliders, options, calculators, date pickers, and several elements to interact with instead of typing. We’re in the DOS era of the GPTs. We’ll see a lot more improvements. That’s my feeling.

1 Like

@barizon - Have you built actions as part of your custom GPT in the GPT Store or is your functionality available via a stand alone application which uses the APIs and function/tool calling?

I have agents with dozens of actions that call APIs and do other stuff. I don’t have standalone applications at all! I don’t need to code. People don’t understand that the paradigm of Forms/Menus/Reports is gone. There’s no need anymore for standalone applications, you need is a multimodal prompt and a multimodal canvas, and that’s it.