Mark (2024-09-21). What Happens When the Free Trial Period of GenAI Is Over?.
Based on how unprofitable they are, I hypothesize that if OpenAI or Anthropic charged prices closer to their actual costs, there would be a ten-to-a-hundred-times increase in the price of API calls, though it’s impossible to say how much without the actual numbers.
This is a valid point; if the companies building frontier models charged users what it cost to run those models, there’s no way we’d pay.
However, I think it makes sense for them to keep providing access at a huge loss, for a few reasons.
- Most of the people at these companies really (really, really, really) believe that the true value of generative AI lies in the future and if they succeed in building AGI, any short-term profits they missed out on will be completely irrelevant.
- No-one really knows a lot about the capabilities of new frontier models. So it makes sense to make them available and have smart people around the world experiment as a way to surface their utility.
- Related to this, companies use our preferences and behaviours to better understand specific use-cases, which they can use to build out more features. Even if they don’t use any of our prompts and data that we upload, they can learn a lot from how each of us uses their models.
- The more value we get from using language models, the more likely it is that companies achieve market dominance and user lock-in. At the moment this is less of an issue because my prompts mostly work, no matter what model I use. The cost of switching is insignificant. However, as companies build more commercial features (e.g. Canvas in ChatGPT, or Projects and Artifacts in Claude), it becomes harder to move your entire workflow (which is premised on those features) across the other models.
- Companies need society to come to terms with what it means to have generative AI integrated into all aspects of our lives. The more we use it – and come to rely on it now – the more likely it is that AI becomes a normal part of life.
- Related to this, free and low-cost access means that developers and other companies can build a thriving ecosystem around the frontier models. The long-term potential for revenue generation is potentially significant. For example, imagine if OpenAI’s models become the foundation for NHS systems.
As you can see, it makes sense for companies to make this technology available for free, even if it means they’re making eye-watering losses right now.
That’s not to say that won’t cut off access at some point, or even start charging us the real-world costs of using the models. But I’d be very surprised if they do this. There is simply too much value being returned to these companies through making them widely accessible.