I need to make sure my Pickaxe doesn't make stuff up

I’ve built a Pickaxe that uses documents to answer user’s questions. It works pretty well, but often when it doesn’t know the answer, instead of telling me it doesn’t know, it just makes stuff up.

Any advice to stop this kind of thing from happening?

1 Like

You just need to mention that you don’t want it making stuff up in your prompt. That should be enough.

But you already knew all that didn’t you :joy:

1 Like

I studied dreams for more than 40 years, read countless books and kept a dream diary all that time. What the AI is doing is producing “suggestions” and that is all they are. The suggestions are informative enough to encourage a more elaborate interpretation and are in no way definitive in meaning. It would help to get a dictionary of “spiritual” dream symbols as all dreams are spiritually connected to the dreamer. Also there are dream dictionaries on-line that define symbols and even symbols with spiritual meanings.
IT REALLY PAYS to interpret dreams, as they protect, encourage and inform. However, dreams have no concept of “time.” E.g., I’ve had dreams come true in a few days, weeks, months and even years. The longest so far came true after 30 years and involved a large sum of money down to the very denomination given in the dream.
30% of dreams are usually physical health dreams.
You should get a LIST of the various types of dreams that occur, it will help in making an accurate interpretation.
It’s worth the effort.
ONE LAST VERY IMPORTANT ISSUE: IT HELPS TREMENDOUSLY TO WRITE DOWN THE “FEELING” THE DREAM LEFT YOU WITH UPON WAKING. YOU CAN THEN PLACE IT IN THE BROWSER BOX THAT ASKS HOW YOU’VE FELT THE LAST WEEK OR SO. THIS WORKS WELL AND REALLY HELPS YOU GET A STEP UP ON SUGGESTIONS.

What you are pointing to are commonly called “Hallucinations”. There are several ways to combat hallucinations. While they are effective, they are not 100% full-proof, so be sure to always check your Pickaxe in the control panel and see that is answering well.

  1. In your prompt, tell it to double check answers for accuracy. While silly, this actually helps.
  2. Specifically instruct your Pickaxe to not make up answers. Include in your prompt, some text like: “If you don’t know the answer to this question, DO NOT MAKE UP an answer. Instead admit that you don’t know.”
  3. If you’ve provided your pickaxe with documents in the form of a knowledge base, go under “advanced settings”. Here you can include some specific language about how the AI should only draw from the information in your knowledge base. (see picture)

1 Like

@jbarmann19 I think there’s a lot of overlap between the realm of LLMs and the Jungian realm of the collective unconscious. I made a dream interpreter bot a few months back—let me know what you think!

Inck: I ran several dreams through your Dream Bot and it definitely serves a purpose and can be used to a person’s advantage if the person is familiar with dream language which include: symbols, signs, signets, emblems and numbers. Which would allow them to use your product to “enhance” their own search for an interpretation.

One shortcoming of what I’ve seen is there should be a way to enter what the dreamer felt upon waking. This is central to any dream as what it does is point the dreamer to what the specific dream would be dealing with. It would also help the AI Program if there was a way to put it to work within the structure of the program itself.
Dreams often deal with numerology or what the average person simply recognizes as numbers. Numerology being what is taught as “fine numbers” which is simply any number reduced to its finest outcome: e.g., 343 would be 3+5+3 = 11, which the 1+1 of eleven would equal the number 2. This is what is used as “fine numbers” in metaphysics. Personally I am not a big fan of numerology unless it deals with dreams. Then for myself it becomes very useful.
What your Dream Bot doesn’t seem to recognize, and which could simply be posted somewhere for others to consider are the various TYPES of dreams. My own list of Dream categories reads: (a) Lesson dreams; (b) Peak Spiritual Dreams; (c:) Contact with deceased individual dreams; (d) Dreams of Power over lower self and base emotions; (e) Lucid dreams; (f) Astral travel dreams; (g) Physical health dreams; (h) Past life memories dream; Premonition dreams.

In other words, what I’m trying to explain is that your Dream Bot is fine, however, to be more effective any novice seeking an accurate meaning needs some type of fundamental understanding of dreams and the basics. This can just be a page people can read, so they can better apply what the AI is telling them.

In conclusion, the two things I would recommend to improve your readership and popularity is to instead of asking what a person’s feelings or emotions were for the past week, rather ask what they felt upon waking from the dream —which as I mention, this is a biggie. The other is simply my recommendation of a page people can read explaining the fundamentals of dreams. Or even a Dream Chart which can correlate AI information with dream categories, or perhaps even numbers and colors. Along with some simple common sense suggestions it would only serve to enhance your website, in my opinion.

While your AI Program is good for people like myself with decades of personal experience with dreams, the novice could easily lose interest because of not understanding the value of the information provided.
I hope this helps, and thanks for providing such a service!
—Jeff Barmann, AA, BSc. Retired Paralegal.

Typo 343 meant to be 353.

Well hey, you can clone my Dream Pickaxe and make your own version by clicking the clone button in the upper right:
image

why in the world this was ever named a hallucination is beyond me. they should be (and ARE) confabulations.
I always thought calling them hallucinations was harmful, both to the mental health community and to the repuation of language models themselves…
:woman_shrugging:t4: