3.5.1 Expensive GPT Assistants
At the time of writing, OpenAI tracks the resource usage of your ChatGPT account, including that used for GPTs. If you hit a resource usage limit and get blocked, your ChatGPT account will also be blocked. Blockages typically only last a couple of hours, but this can undoubtedly be more than a little annoying.
Therefore, we want to ensure that users using your GPT don’t exceed their resource usage limits for regular use. Following is a list of features that increase resource usage while using the GPT:
- Creating images—Image generation is still a premium service, and successive image generation can quickly get your user blocked. It’s generally recommended that you inform your users of the potential risks and/or try to reduce how frequently images are generated.
- Code interpretation—This feature allows for file uploads and running of code for data analysis. If you think your users will require constant use of the coding tool, then inform them of the risk.
- Vision, describing images—If you’re building an assistant that uses vision to describe and extract information from the image, plan to use it sparingly.
- File uploads—If your GPT uses a lot of files or allows you to upload several files, this may cause blocks. As always, guide the user away from anything preventing them from enjoying your GPT.
NOTE Moore’s Law states that computers will double in power every two years while costing half as much. LLMs are now doubling in power about every six months from optimization and increasing GPU power. This, combined with the cost being reduced by at least half in the same period, likely means current resource limits on vision and image-generation models won’t be considered. However, services such as code interpretation and file uploads will likely remain the same.
Making your assistant aware of resource usage can be as simple as adding the rule shown in listing 3.18 to the assistant’s instructions. The instructions can be just a statement relaying the warning to the user and making the assistant aware. You could even ask the assistant to limit its usage of certain features.
Listing 3.18: Resource Usage Rule Example
RULE:
When generating images, ensure the user is aware that creating multiple images quickly could temporarily block their account.
Guiding your assistant to be more resource conscious in the end makes your assistant more usable. It also helps prevent angry users who unknowingly get blocked using your assistant. This may be important if you plan on releasing your GPT, but before that, let’s investigate the economics in the next section.
3.5.2 Understanding the Economics of GPTs
Upon the release of GPT Assistants and the GPT Store, OpenAI announced the potential for a future profit-sharing program for those who published GPTs. While we’re still waiting to hear more about this program, many have speculated what this may look like.
Some have suggested the store may return only 10% to 20% of profits to the builders. This is far less than the percentage on other app platforms but requires much less technical knowledge and fewer resources. The GPT Store is flooded with essentially free assistants, provided you have a Plus subscription, but that may change in the future. Regardless, there are also several reasons why you may want to build public GPTs:
- Personal portfolio—Perhaps you want to demonstrate your knowledge of prompt engineering or your ability to build the next wave of AI applications. Having a few GPTs in the GPT Store can help demonstrate your knowledge and ability to create useful AI applications.
- Knowledge and experience—If you have in-depth knowledge of a subject or topic, this can be a great way to package that as an assistant. These types of assistants will vary in popularity based on your area of expertise.
- Cross-marketing and commercial tie-in—This is becoming more common in the Store and provides companies the ability to lead customers using an assistant. As companies integrate more AI, this will certainly be more common.
- Helpful assistant to your product/service—Not all companies or organizations can sustain the cost of hosting chatbots. While consuming assistants is currently limited to ChatGPT subscribers, they will likely be more accessible in the future. This may mean having GPTs for everything, perhaps like the internet’s early days where every company rushed to build a web presence.
While the current form of the GPT Store is for ChatGPT subscribers, if the current trend with OpenAI continues, we’ll likely see a fully public GPT Store. Public GPTs have the potential to disrupt the way we search, investigate products and services, and consume the internet. In the last section of this chapter, we’ll examine how to publish a GPT and some important considerations.