Four things I wish everyone knew about AI

Four things about ChatGPT and other conversational AI solutions that I wish everyone knew.

Everyone and their grandmother is chatting with AI - which is great! But while it's important to see the possibilities, it's also necessary to understand the limitations.

Here is what I wish everyone knew:

1. Chat models are not databases or Google on steroids. Even though they hold a lot of information, they work by recognizing patterns and generating answers rather than looking up exact sources.
2. Chat models cannot create knowledge out of thin air. They can generate incredible and even innovative things based on the patterns they have seen, but it's all based on what already exists in the training data.
3. All AI systems have a "knowledge cutoff". For example, ChatGPT's knowledge currently only goes up to October 2023 and Claude up to July 2024. Therefore, there is limited information on recent events or developments. Some systems can search the web and provide answers based on that, but the result depends on the sources found.
4. Privacy is not enabled by default. Your conversations can be used for further development of the AI. Be mindful of data security - especially when dealing with sensitive data (e.g., you have to specifically opt-out of data being used for training in ChatGPT).

Often, people are not aware of this - like when an office worker runs contracts through ChatGPT to get a summary, or a student takes answers as the absolute truth without checking sources (both real, recent examples). It is possible to tackle these points with the right approach when building AI solutions, but please keep in mind that this is tech, not magic. 🪄 ✨