gencartabeta

gencarta is an exploratory project I started in the spring of 2023. The goal is to give users a simple interface to explore the vast amount of information captured by large language models (LLMs) .

Chat interfaces, such as ChatGPT, give users an enourmous amount of power since virtually any question/task can be expressed in a text prompt. However, getting the answer you want requires writing an initial prompt and following-up with several more prompts until you get what you want. This makes chat interfaces well-suited for solving a wide range of problems, as long as you can clearly express what you want in a chat.

With gencarta, users enter a small search query and the AI will do the rest of the work. It uses individualized settings and various presentation formats to display the response from LLMs in an interesting, interactive and engaging way. gencarta generates a set of ChatGPT prompts behind-the-scenes and creates a personalized response according to the settings chosen by the user. The response will include text, images, and hyperlinks to concepts that would be worthy of exploration on their own.

This tool does not replace ChatGPT, Wikipedia, or Google. Instead, it fills a unique niche that captures our desire to find information about a topic quickly, synthesized in a way that fits our unique way of learning.

Of course gencarta is not perfect. The main limitations of this tool are also core limitations of all LLMs, in their current form:

• Factual Innaccuracy - Currently LLMs have a hallucination problem. This means they can sometimes generate factually incorrect information.
• Stale Data - Typically LLMs are trained on a static dataset which means they do not know about current events or recent information.
• Potential Bias - LLMs are trained on large textual datasets from the web. Any biases in these datasets will likely be present in the responses of the LLMs.

With these limitations in mind we should use this technology with caution.

I encourage you to try out gencarta and share any feedback you have with me on twitter.

-Germán Adrián (@_driangle)