The models behind GPT chatbots will continue to improve, but they are good enough today. Everything we discussed in the previous sections (1, 2, 3, 4) is possible with the technology that exists today. Model capabilities and personality do play a role in retaining users on a given platform. Trying to win the fight for users purely on your model is a tough fight, and you can easily lose everything if a competitor makes a research breakthrough. Differentiation through model capability doesn’t make financial sense. Models are already becoming a commodity, and this trend will likely continue as research chases diminishing returns.
Providers have work to do to make it harder for users to leave. The personalization of the experience, or even as far as relationship building between AI and the user, can help. In extremes, users are already falling for their AI companions. On a broader scale, creating a comfortable and familiar experience that is personal to each user will make many want to stay. I am not sure if this will be enough to create user loyalty to a particular platform.
If we have an environment where users can easily switch platforms without friction, that may help keep some of the more extreme advertising and data monetization methods in check. Competition is good for end users. Right now, there are enough major players in the GenAI chatbot game that it will be difficult for any single provider to dominate the market. Most of the monetization opportunities we discussed make the user experience worse. Adopting these strategies in the current market is a dangerous game if your competitors do not also take the same steps. Your users can easily move to the least adsy platform at any given moment.
Long-term, it is hard to know what will happen. The field of competition may shrink. Some companies may run out of money or consolidate. Some might decide to focus on AI within their core business and leave the general chatbot race. The fewer players there are in the game, the easier it is to collude, monopolize, and monetize users.
Last, but certainly not least, I want to mention the concerns around privacy, security, and political influence. Current ad platforms based on top of web search have access to a lot of information about you. Many people feel that their right to privacy is already violated by existing data harvesters.
The scenarios we discussed in previous sections increased the details of data collected and enable more fine-grained targeting. Depending on how much detail is shared, adversarial users could exploit the ad framework for harm. Think about a stalker using details about a victim to inject harassing content via targeted ad placement. Even with privacy-preserving data, this type of attack is still possible.
Even without detailed targeting, other types of malicious attacks are possible. Fraud schemes embedded as GenAI platform ads may be difficult or impossible for end users to detect. Political groups could pay to insert their viewpoints into chat sessions on particular topics via paid ads. Sure, content moderation could help, but this hasn’t exactly been an area of investment for online platform providers or a success story of keeping users safe.
The pressure of competition, coupled with the ease of switching platforms, will hopefully keep the current experience tilted in favor of users. One way or another, we are going to have to pay to support GenAI. The test will be whether users will tolerate the true costs of GenAI via paid subscriptions, ad consumption, or both