What to do as a result:
Use AI tools to challenge your ideas.
Tell me why [your idea] might be a bad idea. Give me examples of people or companies who have tried it before without success.
⚠️ AI can be iffy on details
The goal of these models is to get you one great response to what you ask for. One of the problems right now is when a model doesn't have enough information to get you an accurate response, sometimes it just... makes things up.
This phenomenon - "hallucination" - is something AI researchers are constantly working on improving.
This type of inaccurate response does not represent the majority of responses from models and I don't want to discourage you from using AI tools. Most of the responses I get in personal use are accurate.
I expect this to be one of the areas where we see constant improvement in the models as people use them more, and builders create better safeguards.
What to do as a result:
First, check your work.
No prompts here. Just think critically about the responses you're getting and double check important facts against other sources.
If you are looking for a prompt, one of the first things I'll do is ask a "yes" or "no" question about a response I'm unsure of to gut check.
For example, I asked AI to create a blurb about my accomplishments as an artist (something there is very little, if any information online about):
Write a 1 sentence blurb promoting Chris Vasquez, the
digital artist. Include one of his major accomplishments.
The response was flattering, but inaccurate.
Chris Vasquez, the talented digital artist whose work has been featured in renowned art galleries worldwide and created the mesmerizing digital installation "Neon Forest" that explores the intersection of nature and technology.
I'm pretty sure I never created a mesmerizing installation of that name, so I asked:
Did
Chris Vasquez create a digital installation called "Neon Forest"?
And got the response:
I could not find any information about a digital installation called "Neon Forest" created by a person named Chris Vasquez who is an illustrator or digital artist.
Second, make your prompts specific. The more specific your ask, the less room there is for a model to go off script.
Instead of writing overly simple asks like write a paragraph promoting my copywriting services, include relevant information that you have and know to be true to get better results. Something like:
Write a paragraph promoting my copywriting services. My name is Chris Vasquez and I have experience writing emails, blog posts that interest large amounts of people. I also have experience writing technical documents.
👿 AI is bad at generating
completely novel stories
Like I mentioned earlier AI tools give human sounding responses by referencing their training material to give a best guess as to what an answer might look like. Generative AI tools don’t have novel thoughts.
This means that even models that are connected to the internet are referencing information that other people have created, and aren’t independently experimenting and learning new things in the same way humans do.
What to do as a result:
Experiment all the time, and then share your results.