Finish AI Projects: 4 Proven Moves to Overcome Obstacles

Depending on what model you are using, you might find yourself fighting a really neurotic AI coworker. I often find myself wishing Claude would stop telling me to go to bed, or to do 3 checks before going to the next step. It is more often than not because I didn’t give it full context of what was in my head, likely a problem of my brain always going in a million directions and not using the speech-to-text way of prompting that would allow me to conversationally provide context and input. finish AI projects

However, when you inevitably run into a wall in a project, you can do one of several things.

This approach helps you finish AI projects.

  1. Do exactly as the AI asks for you to resolve it.
    1. This could be really good advice, but before you do ensure you are on the same page as the AI as it is only as good as the instructions you gave it in the last prompt.
  2. Have another LLM do the action you are being asked to perform. You can copy paste and say – answer this as if you were me “[request]”
  3. Reframe your question or intention so the AI doesn’t need your input to accomplish the goal. If you expected it to come up with an answer and it has more questions, you probably didn’t give it enough context. As you move up in the intelligence of the model you use, the less you have to specifically direct it on your expectations. You provide it your intent, and your desired outcome and let it come up with the how, stop telling it the intermediate steps that you think it should tackle.
    1. I recently heard about the bitter lesson written, in ancient history all the way back in 2019. It basically says despite humans efforts to make AI more efficient by acting more human, it does better at any specific job by just throwing more compute power at the task. The bitter part is that we expect a human-centric approach to be the best path for training a new semi-conscious entity. We love to anthropomorphize, and yet the bitter reality is our way of thinking for an AI model does not yield optimal results.
  4. Revise the project or skill you are working in to more independently come up with a result you intended without further rounds of back-and-forth. If you are doing something often enough, there should be a pretty repeatable pattern that you can educate an AI to predict and assist. If it needs a lot of input for accomplishing that task, spend some time fixing the system, not the one-off task.

I wrote this post by hand, in full — the internal links above were the only parts AI touched.

Leave a Comment