OpenAI's Grand Experiment in AI Proliferation

OpenAI released its new product, CustomGPT, last week. According to OpenAI, Custom GPTs, aka GPTs, are independent agents/chatbots that are fine-tuned for specific applications. All you need is to tell the GPTbuilder what you want, and the builder will make it for you! They also introduced the GPT store, an app store-like distribution channel for the GPTs. This means anyone with zero knowledge of coding can now generate their own chatbots and distribute them to the public, which can be both exciting and scary!?The exciting part is the fact that this is going to be another revolutionary step towards democratizing technology and knowledge, which, by the way, can make OpenAI one of the most powerful platforms on the internet and a Monopoly! It is similar to what Apple did for developers or small software developing companies almost two decades ago with the AppStore, or what social media like Instagram and TikTok did more recently for content generators, influencers, and brands. The difference is that the cost of generating and running GPTs for the public is almost equal to zero. In fact, I just created a perfect TA for one of my classes in less than a minute, and it works pretty fine. Generating an app to distribute on the App Store still requires coding knowledge or a team of developers, and managing a social media account to grow an audience and create content is effortful. According to a recent study, top TikTok creators spend an average of 8-10 hours creating each video. That's on top of the time spent managing the account, interacting with followers, etc. Whereas with GPT, you just upload your content and prompts, and the automated agent can handle everything else for you. It's fascinating that the model can potentially self-train upon interacting with your audience and self-maintain to update and generate new ideas and content, besides what you originally provided. Now, the scary part comes with the fact that, like Google and Wikipedia, the Free Encyclopedia once became the main sources for public knowledge and learning, AI tools like ChatGPT and custom GPTs risk becoming the new blindly trusted sources for information and truth. With the marginal cost of generating and distributing content nearing zero, the technology has made proliferating misinformation incredibly easy. This is like a genie that, once released, can never be put back in the box. Aside from this, the process of feedback and data exchange between custom GPT agents and ChatGPT is yet unknown to the public. OpenAI claims there is no dynamic feedback, which is reassuring if true. However, misinformation distributed by custom GPTs could theoretically impact the foundational model indirectly over time and by feeding the model with new information. But, I think for now, we have to hope AI developers are aware of these dangers and take steps like using trusted data sources, applying content filters, and monitoring data quality when training new generations of models.