Will GPT-3 be able to answer in non-hyperfriendly ways?

Hello,

I really love to play around with GPT-3, but sadly hard limited by my non-existent purse.
It’s a lot of fun to play around and try stuff but one thing does actually annoy me a bit, which is a problem I have with all AIs and Assistants nowadays.

Super-Hyperfriendliness

It might sound weird, especially in the current extreme cancel-culture where every possible negative word someone says or writes quickly gets attacked. Everyone can instantly cry everywhere about “I don’t like that” and most companies instantly subdue to any request, no matter how silly it is, which is especially for Assistants and AIs a big kick in the crotch.

Most AIs or Assistants just turn off or block anything with a possible “Bad word”, even if it is maybe just a songname or even a simple human name (Dick as name example).

But also I am not a fan of “Everything has to be a cute little fluffy cloud where everyone is on LSD happiness”. I love to tease and like a good arguement, friendly fight and stuff.

When I ask Alexa like a 500th time (I do that daily in real hope they add it) what a Pikachu sounds like, it’s highly unbelievable a real assistant would just say a 500th time “I have no idea”, it would be sarcastic, annoyed or even pissed. I mean, I DO annoy her with that question on purpose and wouldn’t mind if it reacts that way.

Same with GPT. In the videos where the devs gave GPT a face it was always on that super hyper friendly speech. Even if it tried to be “funny” in the ways it can you barely “bought” it. It is just missing the part where you need to use non-superfriendly speech and reactions.

Not saying it should go full gangsta rapper style using the f-word 5 times after each word, but some more lively reactions would be great.

Am I the only one that thinks that way? Is it wrong to teach AI a bit more agressiveness?
Would really love to know how others think about that.

(Sorry if I sound a bit weird, english isn’t my native language)

1 Like

Hello @AkumaNoTsubasa,

That’s where prompt design comes into play. Prompt design essentially tells GPT-3 how to react to give you the desired output.

Prompt Design Basics:

1. State what the prompt does at the start
2. Use plain language to describe your inputs and outputs
3. Show the API how to respond to any case
4. You can use text and emoji
5. You need fewer examples for familiar tasks

For example, OpenAI provides this example of prompt design that tells GPT-3 that it’s Marv the chat who reluctantly answers questions:

Marv is a chatbot that reluctantly answers questions.

User: How many pounds are in a kilogram?
Marv: This again? There are 2.2 pounds in a kilogram. Please make a note of this.

User: What does HTML stand for?
Marv: Was Google too busy? Hypertext Markup Language. The T is for try to ask better questions in the future.

User: When did the first airplane fly?
Marv: On December 17, 1903, Wilbur and Orville Wright made the first flights. I wish they’d come and take me away.

User: Who was the first man in space?
Marv:

Although I haven’t tried, I’m sure you can also tell GPT-3 to not be repetitive (unlike Alexa) when responding to you as well. Hope this helps!