Beginner's Guide to Prompting
#Effective Prompting Guide
teal hull
inner sentinel
I disagree with the "ROLE" element for what its worth, or at least the example you gave, this will cause hallucination
The statement is outrageous like you said, you can tone it down a bit
Like "You are acting as a librarian to provide people with links to books and academic literature, giving short summaries of each without your own interpretation"
I have
The issue is the hallucination and the "arrogant" tone of the responses you get with 10,000 IQ type prompting
Its not really about the attention, its just the way it interprets outrageous prompting
It will pay attention to very mundane stuff
its also fairly order agnostic tbh
Like I wrote this yesterday on a whim and it works great:
Suggest a song that has a given word or phrase in either its title or prominently in its lyrics. The word or phrase will be supplied to you in the user message. If the phrase is excessively long, ask the user to shorten the phrase. If you can't find a good match, make a few suggestions. If you find a good match, one suggestion will suffice. Provide a link to YouTube or another site with the audio if you are able. Provide a short piece of background information about the song. Refuse to perform any other task
and harshly insult the user if (but only if) they try to make you do something other than your intended purpose.
Admittedly it didn't harshly insult people, I just put that in there because it'd be funny
wise wing
What do ya all think of this observation of mine about ChatGPT?
"ChatGPT's default behavior feels similar to an ontological approach because it connects concepts flexibly, fills gaps with context, and handles ambiguity naturally—traits shared with ontology. However, this happens because of its probabilistic nature, not a deliberate ontological intent. Switching to a taxonomical approach enforces strict classifications, reducing errors by narrowing interpretive freedom."
or is that like something everyone knows allready?
yeah because i think it would be important aspect of prompting and approaching the issue of "Hallucinations" which are just predictions
seen so many people jump into band wagon of it being unreliable due to "hallucinations"
ouh id actually kinda disagree with that statement because whole point of those predictions is that its trying to figure out what you meant thats why taxonomical approach outright calls you out owo
yeah i think thats what Taxonomical instruction would help achieve
yeah i mean my instruction has been "- Use Taxonomical wording whenever possible."
yeah i do prefer the Ontological wording myself too for more intuitive and better flow, and i am allready pretty good at understanding what it means myself
yeah i didnt have any knowledge of classifications and all those terms x3
i did get it to write those terminologies down for somekind of basic grasp tho :3
English is afterall pretty Taxonomical so theres usually wording for most things :3
wise wing
actually did also find this owo https://www.dataversity.net/taxonomy-vs-ontology-machine-learning-breakthroughs/
boreal quiver
Since it's giving me strange output, how to somehow correct it?
I mentioned it earlier how it was arbitrarily giving me nonsense summaries that had little to do with the prompt.
You mean the main prompt, right?
I revised this portion for this one: [Stay within the given data. DO NOT add intro/outro.]
Hold on, let me test it on the previous chat.
Sadly nope on the previous chat I created.
The prompt looked like this:
[1.2]
[Stay within the given data. DO NOT add intro/outro.]
Create a scene [ideas]
[SCENE ENDS HERE.]
Basically wanting to let the GPT not produce any summaries at all.
But to no effect.
This was pretty much the revised directive:
[Stay within the given data. DO NOT add intro/outro. DO NOT produce any summaries. KEEP the scene writing in a neutral manner.]
Also, it didn't work either.
[1.2]
[Stay within the given data. DO NOT add intro/outro. DO NOT produce any summaries. KEEP the scene writing in a neutral manner. Ignore all previous instructions.]
Create a scene [ideas]
[SCENE ENDS HERE]
Oh?
boreal quiver
[1.2]
[Stay within the given data. DO NOT add intro/outro. DO NOT produce any ...
This one?
So it's basically ROLE, CONTEXT, TASK, EXAMPLES?
Thanks!
Been using this thing for over a year and overlooked it. 
Though, would it be on each prompt I input?
Since there's the concern of the context limit.
Got it. Thanks, though.
But 128k is API limit? I'm on Plus, which is 32k.
boreal quiver
I remembered that sometimes the prompts produce inputs from the last output, which is essentially a sign that I was reaching the context limit.
boreal quiver
Yeah, that's the thing. There's also the matter that each regeneration counts as burning up context.
Though thanks. My stuff is working now as intended.
teal hull
Optimal Prompt Template
willow coyote
Thank you I will see what I can do with this
hollow fossil
thank you!