Ai Whispering 101.4

Yoooooooo...hello there my LLM Lovers! How are we all feeling today? I hope you’re feeling as good as I am considering that this will be the fourth and final installment to my AI whispering blog (in case you missed the earlier ones, here are the links to  AI Whispering 101.1 / 101.2 or 101.3).

It’s been a long time coming, but the reality is I wanted to make sure I truly understood all of these tips, and therefore set up my own tests of all of these terms… my interpretation of them, and describing them to you makes sense of the dense, heavy reading I had to go through to understand it.

ProzessTec has also been installing local versions of these LLM’s to better support our clients… using public chat based LLM’s, you risk your business’s intellectual property… using local LLM’s, you minimize (maybe even eliminate)  that risk.

Without further adieu, Here is the last installment!

#20: Use output primers, which involve concluding your prompt with the beginning of the desired output. Utilize output primers by ending your prompt with the start of the anticipated response.

This particular prompt tip helps with something called the “warrant” of the request. I am using a university level linguistics (Toulmin’s argument model specifically)terminology (Warrant) to help illustrate the point. A linguistics warrant is essentially the focus of a sentence. If you have ever been in a conversation where a sentence can logically mean to distinct different things, you will realize how important the warrant is. We humans have a great ability to discern the proper focus of the sentence by understanding the entire context of the communication…but LLM’s are not as gifted as we are.

Take this sentence:

Which port are you using?

If we are cooking, we could be talking about a delicious wine beverage…if we are in a data center it hopefully will not mean a delicious wine beverage…lol. We as humans can more easily discern the true intended focus (warrant) the request due to 1 million other factors that provide context… In this very simple case, the context is whether or not we’re in the kitchen versus the data center. This is information that an LLM will not be privy to. So we can help Provide that perspective by helping to start the response…

To the prompt above, we can write it as…

Which port are you using? The answer should begin with, “ the network appliance port we are using is…”

By starting the answer, you help provide that context that LLM’s cannot discern as well as humans can.

 

#21: To write an essay /text /paragraph /article or any type of text that should be detailed: “Write a detailed [essay/text /paragraph] for me on [topic] in detail by adding all the information necessary”.

This one is pretty much self descriptive… It’s essentially detailing how specific you need to get with the LLM to get the desire output. If you want an essay, call it out.  IF you want to paragraph call it out. If you want to blog it out. If you want an ad call it out… it is that you’re looking for it’s helpful to be descriptive as to the format so that the LLM can give you a response that’s pertinent.

 

#22: To correct/change specific text without changing its style: “Try to revise every paragraph sent by users. You should only improve the user’s grammar and vocabulary and make sure it sounds natural. You should not change the writing style, such as making a formal paragraph casual”.

Another easy to understand prompt tip, another prompt tip to be as clear as possible, when prompting the LLM. Remember that the LLM’s Memory is not as good as a human memory. While it can remember some context, every prompt has an ability to generate something new. You can test this out by asking the same prompt twice in a row in different chats. There’s a good possibility that you will get two slightly different answers from the exact same prompt.

 

#23: When you have a complex coding prompt that may be in different files: “From now and on whenever you generate code that spans more than one file, generate a [programming language ] script that can be run to automatically create the specified files or make changes to existing files to insert the generated code. [your question]”

This is a tactical prompt tip advising the LLM specifically to create separate files for the output. Nothing more to see here kids… do not pass “go”, don’t collect 200.

 

#24: When you want to initiate or continue a text using specific words, phrases, or sentences, utilize the following prompt: ‘ I’m providing you with the beginning [song lyrics/story/paragraph/essay...]: [Insert lyrics/words/sentence]’. Finish it based on the words provided. Keep the flow consistent.

Another tactical prompt that will help establish flow and writing style… You are essentially asking the LLM to take your example and continue the rest of the response in a similar way.

 

#25: Clearly state the requirements that the model must follow in order to produce content, in the form of the keywords, regulations, hint, or instructions

This one is slightly related to the warrant prompt tip above. It’s providing better context For the LM to produce a response that  is applicable to the prompt.

 

#26: To write any text, such as an essay or paragraph, that is intended to be similar to a provided sample, include the following instructions: Please use the same language based on the provided paragraph[/title/text /essay/answer].

This last prompt tip is a tactical prompt tip… In that you are showing the LLM an example of the style and format. You want the response to be in, so it can more accurately match its response to something that you can use.

 

That’s it for now… I hope this series was educational and easy to comprehend. While I can’t guarantee it’s easy enough to understand, I can guarantee that it was much easier than the source material.

I honestly believe that the biggest take away to this entire series is…ANI is still in its infancy. You need to be extremely verbose as if you’re talking to a subject matter expert who has a trouble differentiating nuances. This subject matter expert works best when you are beyond clear in your prompts, When you give it examples of responses, And when you give it, the context that is natural for humans to discern.

LLM’s are great, but they still need a little help.

 

Contact Us to learn more about this topic or share your thoughts!