Hey @Forward ! For simple cases, ChatGPT4 doesn’t need Cheemera at all indeed.
Cheemera is needed only if there are situations that require watertight deduction based on both explicit and implicit assertions made by the rules/principles in question.
Here is a very basic example explaining something that ChatGPT consistently gets wrong, but Cheemera gets it always right.
You can have slightly more complex situations like this…
If A & B are true, then C is true and not D.
If D & E are true and F is not true, then G and A are true.
If E, F and H are true then B and C are not true.
G, C and F cannot be true at the same time.
What happens when C is true and E is not true?
… where each letter is a placeholder for any sentence, such as:
The account holder has been KYC-ed
The person in question has been sanctioned by the OFAC
The contract in question involves a consideration
You have a rule set involving hundreds or thousands of rules (e.g. laws, philosophies, principles), and Cheemera will be able to give you watertight deductions about the implications of any specific situation you would want to inquire about.
Thank you - so basically if I feed it a bunch of company policy rules but then I have a rare situation where some employee did something we did not have a policy to cover, the GPT may not do a good job with that simply because it has to combine rules and use logic?
Or are you only using these methods to test the AI’s ability to solve classroom logic problems?
No there is not a specific problem - I am likely being confusing. I was trying to understand if you use this for specific problems or just to test the AI’s capabilities.
On another note, I have found success with simply asking it to go “step-by-step” and list it’s steps - why is not that a sufficient way to solve for whatever you are trying to solve for here?
Ah ok, gotcha. Cheemera started as a mostly academic experiement but I’m posting it here to see what others could do with it.
About the comparison with ChatGPT4, ChatGPT4 is okay for simple problems, but you do need to double check both the formulation of the problem and its reasoning to be sure that it’s ‘thinking right’. When you are using Cheemera, once the rules/beliefs are properly formulated, you can be sure that the deductions are always predictably and correctly executed.
I can’t believe you just explained that how you explained that. Ok so, it’s a prompt structure that evokes replication of the this specific downstream lines of thought. Nice job!
That’s brilliant. Are there others? I am going to try it out. Thank you!
@gmkung , I read through this thread and thought of potential use cases. Business systems analysis is a potential area.
We can think of any business as a collection of resources and interconnected processes with steps and tasks. Each process has a desired output that ate inputs to other processes. The final output is customer value expressed in goods and services.
Under a productivity strategy, if there is a business problem (effect), the flaw (cause) must be within the quality/quantity of resources or a process breakdown. Similarly, if the company has a growth strategy, an enlargement of resources and/or systwm processes must be employed.
What are your thoughts of such a use case?
A month ago I first posted about Cheemera, a Custom GPT that uses an inference engine external to ChatGPT to perform complex logical deduction, something that the standard ChatGPT struggles with beyond basic use cases due to hallucinations and variability.
I got really good feedback from several community members who reached out, so I went on to improve the Cheemera GPT (now at v1.1) to make it much easier to get useful results. I’ve included a ‘Quirky Animals’ demo in the GPT that features a large number of rules to showcase Cheemera’s ability to handle complexity with ease.
For those building or looking for solutions related to compliance, regulation or risk and needing to manage implications of a large number of rules, I’d love to hear your feedback and see what we can do together
hi @gmkung, I am really amazed, excited and enthralled at what you have done. I have a very specific use case that i have been trying to implement since months now, but have not succeeded.but i see hope in this “Cheemera” that you have given birth to. can we connect to discuss my use case?
I multiplied 2 numbers to find the PRODUCT and also added them to create the SUM, and kept them a secret. Then, I told person A only the PRODUCT and asked him whether he could determine the unique numbers. He said NO, which was also heard by person B (only that A could not find the numbers). Next, I told the SUM only to person B and asked him whether he could determine those numbers. B said “NO,” which was also heard by person A (but not the SUM). Hearing NO from B, A said, “Well, now I can tell you those numbers.”
Question:
Tell me what those numbers are and the reasoning behind it.
