I recently had a conversation with Assistant and found the experience to be very helpful and informative. The AI provided timely and accurate responses to my questions, and I appreciate the effort that went into training and developing the system. One area where I think there is room for improvement is in the AI’s ability to understand and respond to more complex or nuanced queries. Overall, I am very satisfied with my interaction with Assistant and look forward to using the service again in the future.
I find that prompt engineering requires one to actively engage in a theory of mind. Getting an undesirable response and using that to recontextualize your prompt with a follow-up question has had promising results for me. I think of ChatGPT like a feral encyclopedia, and I have to learn how to tame it in order to get the best use from it.
GPT-bases AI does not “understand”; it simply responds based on prompts. The better the prompts, the better the responses.
In the near future, people who are highly skillful at this type of GPT prompting will be highly paid for those skills.
Prompt engineering is already hacked by other AI models. For example, I had a really hard time creating cool looking images with StableDiffusion or Dalle-2. So I went to a prompt creation service on HuggingFace and BOOM everything looked awesome!
Now, in the future, most of these services will go to the wayside, because I believe companies such as OpenAI do not want people to input cryptic AI generated prompts to get good results. But rest assured, people are already getting AI assistance to generate their prompts. Digital steroids. Hopefully in the future this won’t be needed.