Open AI Custom GPT hype is real! Who is really winning?

I don’t think everyone is going to emerge out as a winner from this new OPEN AI custom GPT hype, Not even if you are really good at prompt engineering.

Open AI has really doubled down on their “AI for all” game by introducing the new BOT store, But let’s be honest they have a win-win situation here.

Let me explain how…

First, its only available for premium subscribers, so if you are thinking of creating one, its US $ 20/mo. to get started. For hobbyists and small-scale experimenters, this might not be a trivial expense.

Secondly, to really have a chance to make your GPT’s stand out you need to train these custom GPTs with data that is unique to the GPT model 4 that would really give you edge over other GPTs.

Custom Bots have opened the doors to the world Chat GPT didn’t had access to

Open AI was under question earlier this year and already has a couple of class action lawsuits alleging “unprecedented” and “unethical” web scraping practices.

And the new custom GPT store is going to make it much easier for the company to train its future GPT models on more data that would not have been available otherwise.

So, in essence, they get money and data, a clear win-win for the company.

But what about the rest of us? Although OPEN AI made it sound fairly easy to come up with good prompts to put creative bots out there, the reality is quite different. Even great prompt engineering skills without access to custom data won’t make much of a difference.

If it was as easy as taking prompt engineering skills and creating custom bots, chat GPT would be smart enough to create an army of custom bots covering every foundational existential use case.

GPT 4 is smart enough to engineer its own prompts

Let’s say I wanted to create a custom GPT for a lifestyle coach, first thing it does is suggests me a name for my Bot (see below).

Now, let’s look at configure option of the BOT. And here is the surprise!!

Without adding any additional details about what the BOT does, custom GPT auto engineers its prompt and some conversation starters, which is really cool but again brings back the question is it really gonna be useful if Chat GPT can do the same.

What makes the hype real

The real differentiator is for those with proprietary access to data that GPT 4 has never seen before.

This includes datastores and metastores of organizations, startups and individuals that Open AI has not scrapped yet, atleast directly!!

For example, CNN or any news organization wants to create a custom bot that focuses on delivering latest stories based on user preference, that would really make sense for them to connect a custom BOT to their journalism database.

Now that would be a differentiator and something meaningful!!

The best feature that I believe is really gonna be useful for businesses is to connect their custom bots to their APIs which masks the underlying datastore and gives them benefits of enabling LLMs & Generative-AI features.

This added feature is called “Actions”, where you can connect externalized APIs to retrieve information real-time.

Things to watchout for

One of the issues that caught my attention while training custom BOTs was the lagging performance and delay in response when I connected custom BOT with DALL-E image generation.

The DALL-E model for image generation takes text prompts and generates AI images which really takes quite a bit of time, so if you are thinking of putting a custom BOT out there, watch out for performance when embedding image generation.

Another performance delay I realized was then it called external APIs, now this can really be either on the server side where the API is hosted or network lag on OPEN AI side.

It would be interesting to see when the BOT store goes live if these issues are fixed and BOTs can really serve the purpose without any performance lags.