Ask AI Twice

How to get a decent answer from AI

Reminder: It’s Simulated Intelligence not Artificial Intelligence. It’s software, it’s not intelligent!

One technique to get better results from a large language model (aka AI) is to ask it twice. Don’t use the exact same wording. Rephrase your question slightly and see what differences in response you get.

An example I did myself was asking about discoloration inside potatoes. The first response was that this was perfectly normal and completely safe. The next time I asked it, I mentioned refrigeration and suddenly it was very concerned and telling me about the dangers of discolored potatoes and on the verge of telling me to throw it away and not eat it (the internal discoloration is just fine as long as you cut it out and don’t eat that part).

Simulated Intelligence is extremely helpful for software development. A big factor in this is probably because software development languages have well defined syntax whereas English is very vague and full of contradictions.

Which Simulated Intelligence should you use?

Gab did a very interesting study, asking the same questions to multiple AIs and scoring the bias of left wing vs right wing. This points out very clearly that Simulated Intelligence is not intelligent, it’s simply pattern matching off the information it was trained on. The majority of information published has a left wing bias thus unless you actively work to eliminate this you’ll find AIs overall have a very strong left wing (aka “liberal”) bias.

ChatGPT has the most liberal bias. Gab AI (“Arya”) has the most conservative bias. Grok is the most balanced in their testing.

Most people seem unaware of “Lumo”. Lumo is my favorite simulated intelligence. It has been improving nicely and usually gives a very good balanced view, pointing out both sides of an issue.

Grok is my second favorite simulated intelligence.

Use with care!