Hello, I’m new to GPT-3 and the community, and I’m looking for things to do to push GPT-3 to its limits. Does anyone have any ideas to try? Thank you!
-Syntax, community newcomer.
3 Likes
try and create a chatbot?
2 Likes
You can help me solve the Control Problem.
3 Likes
Well that has already been done multiple times, but I’ll try anyway!
This is amazing! I do see amazing progress in it! Although, ideas are meant to have holes poked in them, so, what if someone hates one thing (E1) but another person likes E1, would the bot go on a majority rule based on person to person, saying if most people dislike E1, then it will not do E1 to anyone, or will it say if most people dislike E1, then it will start off not doing E1 to anyone, but if a user says they like E1, will it do that to them but not anyone else? If so, then how would it remember a certain person likes a certain thing on a large scale, this would be very inefficient with memory, and if you have to pay for the memory, then that will be costly. Do you see my reasoning?
Yes, and it’s important to note that there’s no law saying that you need universal definitions. Thus, as you anticipate, the AGI has the freedom to explore what suffering, prosperity, and understanding means for every individual. There may be some consensus, but no universal definition is required.
1 Like
Ok, I believe if you were to release this into the public it would be important to note that the AI is not using the universal definitions, but it’s own definition of suffering, prosperity, and understanding. This would help with people understanding that it is not using the universal meaning, and would increase the understanding on both parties involved.
No, you’re close but not quite getting it. These are heuristics, meaning that it’s not “it’s own definition” it is meant to learn as it goes, so it can adapt to everyone’s needs individually. If you look at my other work on Raven, I demonstrate that it can keep track of every user’s individual needs.
I would define that as the AI’s definition, the definition it sees for suffering, prosperity, and understanding the AI currently has to me are the AI’s current definition of those words.
But that’s just what I see.
Here is the thread that might help you to get started!
Hello there, check out some creative ideas I collected in this overview: https://towardsdatascience.com/20-creative-things-to-try-out-with-gpt-3-2aacee3e2abf?source=friends_link&sk=15062436a97e320c8796214eb12c39e9
mabye you can make a auto completion bot
2 Likes
Are you using the API? Either way, I would spend a lot of time cleaning and structuring your text. Depending on which model you are using, the words and characters you use will radically influence the information you get in return. For example, using the sentence ‘Hello, I’m a robot.’ compared to ‘Hello, I am a robot.’ will yield quite different tokens. I would also recommend less punctuation compared to more. (So steer away from the amount of commas I’m using).