As we all know, OpenAI brought us a lot of surprises at the Developer Day.
For example, according to Bilibili UP @赛文乔伊， within 24 hours of the release of GPT-4 Turbo, talented individuals from all over the world have already started to showcase their skills:
Some have used GPT to re-narrate Messi's football matches, some have used OpenAI's Vision API combined with TTS technology to commentate on League of Legends games, and some have turned GPT into a game NPC for casual conversations. There are even programs that can generate animated images from natural language descriptions in just a few minutes, and websites that can be built based on hand-drawn sketches.
But is this all that the OpenAI Developer Day is about? It may not be that simple.
Dharmesh Shah, the founder and CTO of HubSpot, a software development company in the United States, recently wrote an article stating that on the surface, GPT-4 Turbo, which was launched at the Developer Day, is the most significant update and the most powerful language model currently available. However, this does not necessarily mean absolute strategic advantage.
Because if we set aside individual models, the underlying interfaces of these language models are actually very consistent. Whether it's GPT-3.5, GPT-4, or Claude, the process of inputting text and obtaining text output is relatively simple, and switching from one model to another is not difficult for users.
This means that the moats between large models are not as wide as they seem, with new language models being introduced almost every week, and their underlying interfaces tending to homogenize. Users can even use multiple models simultaneously or switch between models freely.
However, this is also a subtle but important change at the OpenAI Developer Day.
During the event, OpenAI introduced a series of new features. In addition to improvements to the models themselves (such as capacity, timing, and speed), there are also features that make the models easier to use, enhancing the experience and usability for developers. The higher the abstraction level of these features, the more they are focused on the platform developers.
At the same time, with the introduction of these new features, ChatGPT is no longer just a simple chatbot, and its underlying interface is no longer limited to text input and output.
For example, in the new &#34;Threads&#34; API, OpenAI can manage memory for users, automatically remembering conversation history and saving developers' time, eliminating the need for them to manage conversation history.
If more and more developers start using these new features, they will no longer be able to switch to other models at will. Instead, they will have to consider whether these new models also have similar memory capabilities. Even if the new models do support it, they will need to figure out if any code changes are necessary to match the new models' way of managing memory.
Not only the memory function, but also other APIs such as code interpreter and data analysis are the same. These functions are not only powerful, but also make OpenAI's models increasingly different from others.
Dharmesh believes that as OpenAI's users gradually develop habits, their switching costs will become higher, thus widening OpenAI's moat.
In the long run, OpenAI may become more valuable.

OpenAI

ChatGPT is no longer just a simple chatbot, but an increasingly broad "moat" for OpenAI.

Not only powerful new features, but OpenAI has also taken a "strategic step" forward.