I think my biggest (technical, long-term) issues with current generation AI, primarily through LLMs such as ChatGPT, are the following:

  1. You cannot readily replicate responses, which is the entire point of the scientific process. To ensure correct results for discovery and use is the ability to easily replicate results.
  2. They are primarily cloud based, which means you will not be able to stand them up for testing in 1-5 years and replicate any sort of results or research papers that may be discovered now. Technical, direct, or anything else.
  3. They are primarily cloud based, which means any thing you throw at them, will not be able to be tuned with similar characteristics without massive manual testing as they are internally changed all the time.
  4. They are primarily cloud based, which means you cannot have any ownership or security in the knowledge, that they are safely utilizing your data.
  5. They are primarily cloud based, which means they can price you out of their system at any time.
  6. They are primarily cloud based, which means that they will price you out of their system at any time.
  7. They are primarily cloud based, which means they will break your/their API, and there will be no recompense or ability to change providers.
  8. They are all primarily a private company, and if you build on their platform, then you have no platform of your own. Your business, is dependent on, and is their business.
  9. You own nothing. They own all of it. No matter the legal contract you have with them will protect you from that.
  10. They are primarily cloud based, they are primarily private businesses, and they are anti-scientific development as they are black boxes of inputs, controls, and outputs.

Anything you develop on them now, will not work the same, if at all, in the future. You will have virtually, tangentially, literally, no legacy or agency.