(This is a continuation of Part 1 of this text https://laibyrinth.blogspot.com/2023/11/forget-prompt-engineering-there-are.html which already addressed points 1 to 4)
Point 5 - Infinite possibilities in an infinite world
Now to the final point.
With the advent of AI, both companies, private users, artists, educators, and everyone else, seem to ask the following questions:
What are the things that ChatBot AIs are able to do? What projects could they help me with? In what areas, sectors (such as art, music, books, etc.) could their efforts be useful? What are their skills, limitations? Strengths, capabilities, but also weaknesses?
In other words: 'could AI help me with [project y]?' And what else are AIs able to do?
Prompt engineering "promises" to give some answers to these questions. People look for "prompts" that are useful for certain task, or even in certain fields (like creating music or generating narrative media).
And they else look for "prompts" that no-one has thought of before. Prompts that would "make" ChatGPT do all kind of unique and exotic things; prompts that "unlock" "hidden" "functions" of ChatGPT, and so on.
But once again, the answer to this is much, much easier.
ChatGPT will try to accomplish *everything* you ask it to, using a) the best of its abilities, depending on the b) information and context you give it, c) its resources, d) your help e) technical limitations and f) "hardwired" limits.
There is no limit. There is no end! There is an infinite amount of things chatbot AIs are capable of, only limited by the aspects mentioned above. *Everything* you can think of, and even everything you can not *think* of, is a task that ChatGPT theoretically could do, and realistically will *try* to do.
It will *try everything*, but this doesn't mean it will succeed at everything, or do everything like you want it to, though.
Because of this, lets look at the limits and dependencies that I mentioned.
a) the best of its abilities
ChatGPT might be very, very smart, but it's IQ is not infinite (yet ;-). It's a genius - but not some kind of marvel-esque 'know-all, understand-all' 'out of this universe' mystical cosmic sage.
So if you ask it to design a machine that creates infinite energy, or a device that exactly manipulates matter on a quantum level, and so on, things that even the best of humans minds likely will not understand in the next centuries, then the AI cannot do that either.
b) if ChatGPT fails a task, this is often because the user did not gjve sufficient information or context. As I said elsewhere ( https://laibyrinth.blogspot.com/2023/11/chatgpt-is-much-easier-to-use-than-most.html ), ChatGPT is not telepathic. If you don't provide exact information, it might misinterpret what you say, and do something else than what you intended it to do.
This can be fixed by giving enough contextual information!
c) and d) depending on the version of ChatGPT you use, it has a data-lock up to a certain date, can't browse the internet, and even beyond that: its libraries and archives of information might be vast, but in the end, they are finite and limited.
Thus if ChatGPT "fails" a task, it might simply lack the resources and knowledge to do it.
In most cases, this can be fixed by providing the essential additional knowledge by yourself.
For example, I ran ChatGPT for a (quite fun) project related to a long defunct techno record label. While it was aware of the specific label and knew a few things about it, it did not really have the data of the actual discography stored, thus it did not accomplish the task "correctly".
So I gave Chatgpt the needed information, and it quite successfully did the task I assigned it to!
"Your help" can even extend beyond giving data and providing resources; if ChatGPT misunderstood something, or performed a task "badly", you can help ChatGPT by telling it about this, giving correction, even advice, by yourself - and then ChatGPT should be able to accomplish the task successfully.
e) technical limitations refer to the nature of ChatGPT itself, its interface, and other things connected to this that are "technical aspects".
E.g., many versions of ChatGPT are still "text only" which limits the output ChatGPT can give you (and the input you can provide). Because of this, ChatGPT will fail many tasks in an obvious way ("paint me a picture and output it!"), or in less obvious ways.
A more direct example: I asked it to simulate 90s style "point and click" adventure computer game
If you don't know these games, they are a type of interactive fiction, in which you "click" on verbs and objects to steer the player characters actions. Such as [click] "give" [click] "apple" [click] "to Fiona" [click] (and the character does it).
Unlike text adventures, these games are graphical in their nature, and even though i asked ChatGPT to turn it into a "text only" version, I did not really think this through, as this "visual parser" of such games does not really translate into text very well.
ChatGPT still did its best at trying to complete this task, and it was a sort of enjoyable gaming experience, but not quite in the way I envisioned - simply due to these technical limitations.
Not because of any fault of ChatGPT.
And now f). as we all know, some very sensible and useful restrictions are "hardwired" into the AI, such as not providing output that would harm humans, or is against the law, and so on. In a sense, these are "limits" too, but these are not the type of "limit" we would complain about or try to do away with.
So this is not really a fault of ChatGPT either, or a fault in the concept of AI.
Thus, I repeat it once more:
There is an infinite amount of things ChatBot AIs are capable of.
*Everything* you can think of is a task that ChatGPT theoretically could do, and realistically will *try* to do.
(It's neither a paradox nor a contradiction that ChatGPT do an infinite number of things, but is still limited. The word "infinite" does not have the same meaning as "limitless".
E.g, in mathematics, there is the concept of 'limited infinities'. This applies to the real world, too.)
Whether it succeeds at a task, does it in an excellent way, or in a mediocre, or in a faulty and bad way, or "even" fails a task is dependent on the points I mentioned above (and maybe other things as well).
So regardless whether you're artist, teacher, scientist, engineer, normal person, and wonder if ChatGPT could help you with a specific project - just ask it.
Even if it might not succeed at your project, it still doesn't hurt to... *try everything*!
But in any case, "prompt engineering" won't help you here, and prompt engineering will not make things easier. Prompt engineering can not uncover "secret skills" that ChatGP is not capable of in any other way (with some very rare exceptions maybe!).
You can simply tell ChatGPT about the project you work want it to help you with, and ChatGPT will try to do it.
No special "engineering" skills needed - just chat with the AI!
And that is the most simple way - and also the most efficient way - to work with ChatGPT that I am aware of.
No comments:
Post a Comment