Hacker News Reader

Comment by andyljones | Hacker News Reader

Nic Cammarata on Twitter pointed out that if the prompt gives GPT-3 permission to indicate questions are ridiculous, it'll do so reliably:

https://twitter.com/nicklovescode/status/1284050958977130497

View on HN

More generally, the best commentary I've seen on GPT-3 comes from following @gwern and keeping an eye out for whoever else they frequently interact with.

https://twitter.com/gwern/with_replies

Most other commentary I've seen either goes too far into reactionary skepticism, too far into to-the-moon style hype, or just straight-up gets stuff wrong by not being aware of some technical detail (importance of prompt design, BPE limits).

At least it is witty:

Andrej Karpathy @karpathy

> By posting GPT generated text we’re polluting the data for its future versions

John Carmack @ID_AA_Carmack

> Was just about to say that... Old printed books will be the equivalent of the sunken ships they cut up for uncontaminated steel today.

>Old printed books will be the equivalent of the sunken ships they cut up for uncontaminated steel today

Already are.

What are modern books contaminated with?

> Old printed books will be the equivalent of the sunken ships they cut up for uncontaminated steel today.

It's not so much contamination, but value drift.

Did you ever try reading Shakespeare without commentary? So many things of that language are no longer colloquial. Maybe generated text can recover that and train us.

Or imagine the "dark tongue" from Paul Kingsnorth' "The Wake" to have been real: we've come quite a bit in expressiveness.

That’s true. In general, if you customize the prompt for a specific type of question, GPT-3 will be better able to handle that type of question. Prompt size is limited, though, so if you are attempting to create a general-purpose chatbot out of GPT-3, you won’t be able to handle every archetype of question in the prompt.

Came here to say this. In the prompt there was no example of rejecting a question for being nonsensical.

Humans don't need a prompt to do this.

Children often answer questions nonsensically. They are prompted over years to be serious and answer questions correctly.

I am not sure that's the same.

Children answer nonsensically to normal questions, but if you ask a nonsensical question they are more likely to go "uh?" than to answer with logically built follow-up nonsense.

Yes, they are trained over time, just as this model has been.

What would the ethics be of using children in Turing tests? If you asked a human child to answer Turing test questions, would the judges think of them as a computer?

True. Maybe the human corollary to this is sarcasm. Lots of people fail to pick up on sarcasm in online discussion, hence the /s at the end of a post. And for that matter, natural language parsers really struggle at extracting real meaning from sarcastic texts too.

Humans have much more context at their perusal than a simple prompt.

That depends on context. Nobody would answer "this question makes no sense" in an exam, instead most people assume they lack knowledge of what's being referenced. Even humans have to expect the possibility of nonsensical questions

"None of the above" is sometimes the right answer on exam questions.

So think of this as adding a "none of the above" option.

Humans don’t need AC power to operate either, but presumably these aren’t the kind of differences we’re interested in.

It depends whether you think the other party would prefer for you to play along or call them out on their BS. Between friends we generally prefer the latter.

Thanks for the tip.

I incorporated this into the GPT-3 based search engine I'm hacking right now: https://twitter.com/paraschopra/status/1284905727388028928

What does ‘yo be real’ stand for?

"be real" for being realistic, asking a better question that is not nonsense.

Loosely translated: "yo be real" = "Hey, please be more realistic with your future questions"

Ah sorry I thought it was some technical thing!

"be real" when spoken by humans means "be yourself" not "be realistic".

Depends on the context. But without context I’d read it as “be honest” or “tell the truth” first. So in the context of answering a question it pretty clearly means “ask a question I can realistically answer”.

Out of curiosity how old are you?

It's an Americanism.

ebonics https://www.youtube.com/watch?v=RrZlWw8Di10

Wow, I wonder if it would make a good bullshit detector?

I'm sure I'm not the only one who knows people that can't say "I don't know", to the point they're not far off from the first example.

Reply on news.ycombinator.com