ChatGPT: AI Excellence, Falls Short on Product

Picture the artificial intelligence (AI) from films like Terminator or Ex Machina – autonomous, sentient, and deeply intelligent. Now, allow me to introduce you to OpenAI’s product, ChatGPT. It’s not what you might expect from these film portrayals of “AI”. It’s neither autonomous nor sentient. Not only that, but it’s hardly even intelligent. However, give it the right prompts, and it can mimic intelligence so convincingly that it might outshine many people you’ll meet. It’s like an illusionist, skilfully diverting your attention from the reality of its limitations. But it’s not a good product. This piece attempts to provide an honest exploration of ChatGPT, as a product, a groundbreaking yet imperfect AI chatbot.

Prompts. I mentioned prompts. What is a prompt? Well, by definition, a prompt is an act of encouragement and regarding AI, a prompt is a short piece of text that is used to guide an AI model to generate a specific output. For example, an AI prompt for a ChatGPT might be “Write a poem about love.” The AI model would then use the prompt to generate a poem about love.

AI Model? Ah, yes. ChatGPT, or more accurately, GPT-3.5/GPT-4, the underlying models for ChatGPT, is a Large Language Model (LLM). An AI model is like a brilliant computer that can learn to do things on its own. It does this by being trained on a lot of data, which is like giving it plenty of examples of what to do. Once it’s trained, it can use that data to learn how to do new things. For example, an AI model that’s been trained on a lot of text can learn to generate new text, like poems or stories.

Digressions aside, the process of interacting with ChatGPT involves providing a prompt, to which it responds with an output. For example, “Write me a 500-word article on why ChatGPT is a bad product” and it might produce the piece you’re reading now. 👀 Frankly, it’s brilliant and transformative. The quality of the outputs when giving the right prompts is astounding — it can speed up so much and unlock tons of energy that was spent on the mundane.

However, that’s as far as it goes. Despite demonstrating in November 2022 the vast potential of AI, ChatGPT as a product leaves much to be desired. It’s just “New Chat”, give your input and get your output, your past chats are in a non-searchable list on the side. This setup makes it very hard to automate things through the use of ChatGPT. You can’t save useful prompts on the platform, you don’t get prompt suggestions, nothing. You have to either use an external service, manage a list yourself, or produce every prompt ad-hoc. The product would be magnitudes better if they had prompt templates, allowing for reusable prompts.

Example template

{{context}}
------
Can you check the above paragraph for grammar mistakes, provide a revised paragraph

The above template would take an input, and then ChatGPT would run the predefined prompt, replacing context with the input. This is a simple example, but it is the kind of thing you might run multiple times a day, and it would speed up workflows to have this essentially at a click. Open templates, select one, provide an input and get an output.

To take it further, what if we schedule actions? Using ChatGPT plugins, we could connect a template to a third-party service, and then a prompt could run on a schedule in the background.

I just mentioned plugins, but didn’t say what they are. ChatGPT’s plugins are third-party tools that extend the functionality of ChatGPT. They can be used to access up-to-date information, run computations, or use third-party services. Currently, plugins cater primarily to developers, remaining largely inaccessible to regular users. Again, adding to the narrative that it’s not a good product.

Microsoft and Google’s approaches to AI products are much better, as they are both integrating into existing products. Microsoft’s approach to AI with Copilot is to create a tool that can help people be more productive and creative. Copilot is integrated with Microsoft 365 apps, Teams, Word, etc, so you can use it to help you with your work. For example, you can use Copilot to help you write emails, create presentations, and analyse data. Google is doing something similar with their suite of services.

To conclude, what OpenAI is doing with AI models is brilliant, even Microsoft is using them to power their AI offering. But ChatGPT as a product just isn’t there. There’s so much more they could be doing with the momentum they have. In its current form, ChatGPT is little more than a technology preview. With that said, I will continue to use it daily for the foreseeable future.