If one day, AGI in its strictest sense truly emerges, then without a doubt, this question will be meaningless.
However, that day has not yet arrived, or rather, there is still a visible long journey ahead. Therefore, this question may still hold some significance today; in fact, if one pays close attention to the GPT Store data, some unexpected insights might emerge.
The first image shows the most popular GPTs listed by OpenAI. In the image below, [7] is for generating cartoon avatars, which is more leisure-oriented. [1] is for text-to-image, which could arguably be for entertainment (though most use cases are actually for productivity; for example, when I write articles and need a cover image, I use DALL-E 3). The remaining ten applications are all productivity tools, focusing more on knowledge comprehension and processing. As we know, GPTs users are paid GPT subscribers, and this distribution suggests that at least 70% to 80% of paid users are paying for "productivity."
The second image shows the top 50 GPTs from a third-party statistics site, whatplugin.ai. The reason we need third-party data is because it includes a crucial metric: the number of conversations. This is the most important popularity indicator, supporting the above conclusion from a more objective dimension.
The third image, also from whatplugin.ai, shows the number of GPTs in various fields. While I believe the specific numbers may have discrepancies, the relative comparison is valid. Image generation still ranks first, but as previously mentioned, it serves primarily as a productivity tool by directly "replacing" copyrighted image libraries.
We can almost conclude that for so-called "large models," about 70% to 80% of the user's willingness to pay stems from the "productivity tool" attribute.
AI is expensive—training is expensive, inference is expensive, and it cannot host ads. Currently, the only certain profit models are user subscriptions and API calls. Even OpenAI knows that AI R&D without cash flow is unsustainable, let alone other players.
"Productivity tool" capability is, at least for now, the most important standard for measuring AI large models.
Are productivity tools limited to document processing? Clearly not. They include image generation, video generation, film and television creation, game development, customer service, e-commerce, code generation, and even music generation—there are many.
However, we perhaps need to clarify: for the same function, there is a difference between entertainment and serious use. For the former, no matter how grandly it's described, it may be difficult to generate cash flow and upgrade.