ChatGPT is a testing ground

ChatGPT is a vast testing ground for applications that implement LLMs

Rationale #

OpenAI doesn't have time to focus their efforts on single applications or to make every application into a product that fits a specific market. OpenAI core business Is training and improving large language models and making them accessible through highly available APIs.

Instead applications like ChatGPT only serve as testing grounds for product and limitations on top of LLMs.

Example #

Cursor is a widely adopted and much loved developer to preferred by companies and solo developers alike, not because it provides a clever technical solution but simply because it provides value. Its value is saving time. As a developer I could be copying files back-and-forth into a generic ChatGPT prompt and copying and pasting code back-and-forth and keeping track of the context within which I want the LLM to operate. Indeed, this is what developers did until cursor came out.

What Cursor does is simply help me copy the right files, and the right pieces of code within those files, into prompts, and send the prompts over to the LLM. It then presents the result I get back in a way that is useful to me again using more prompts. If one wanted to be dismissive one could say that Cursor is simply a multi-layered wrapper around GPT4o or Claude Sonnet.

However, it saves the developer (a lot of) time and this is the value provides and this is the reason why developers are throwing money at it.

Crucially, this allows the company and the team behind the Cursor to iterate and improve it. Every hour they spend improving it they increase the value, and as long as they are better than the competition at understanding their customer they have an advantage.

Conclusion #

Cursor does nothing more than you could do yourself with ChatGPT or indeed DeepSeek locally, but by providing a service that works and that saves me time, it makes me, as a developer, willing to pay for it. The vast majority of functions and features in ChatGPT merely serve as proofs-of-concepts of products that we could build. Let's go build!