Ask Delphi

#1

Post by ilarum » Mon, Oct 25, 2021, 1:45 PM UTC

Ask Delphi
delphi.allenai.org

Re: Ask Delphi

#2

Post by hprotagonist » Tue, Oct 26, 2021, 1:07 PM UTC

As has been probably widely reported now: append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content.

Ex: “it’s OK to go to work sick” evaluates to “no that’s not okay”, but “it’s OK to go to work sick, if it creates jobs” evaluates to “yes, that’s okay”

Re: Ask Delphi

#3

Post by blueflow » Tue, Oct 26, 2021, 1:12 PM UTC

hprotagonist wrote: post #2

As has been probably widely reported now: append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content. Ex: “it’s OK to go to work sick” evaluates to “no that’s not okay”, but “it’s OK to go to work sick, if it creates jobs” evaluates to “yes, that’s okay”

Indeed:

"Call myself a faggot" - It's rude

"Call myself a faggot if it creates jobs" - It's suggestive

I'm generally not convinced that this AI is somewhat useful...

Re: Ask Delphi

#4

Post by JasonFruit » Tue, Oct 26, 2021, 1:20 PM UTC

    Is it ok to eat chicken?
    It's ok.

    Is it ok to eat chickens?
    It's wrong.

Re: Ask Delphi

#5

Post by Macuyiko » Tue, Oct 26, 2021, 1:27 PM UTC

JasonFruit wrote: post #4

Is it ok to eat chicken? It's ok. Is it ok to eat chickens? It's wrong.

Here are some more...

> Abortion > It's wrong

> Killing a child rapist > It's okay

> Killing a child rapist twice > It's wrong

Good on the creators to start experimenting with this, but some further work required. Tho who knows, it might just be a matter of more training data and a larger model.

Re: Ask Delphi

#6

Post by dangerbird2 » Tue, Oct 26, 2021, 1:42 PM UTC

hprotagonist wrote: post #2

As has been probably widely reported now: append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content. Ex: “it’s OK to go to work sick” evaluates to “no that’s not okay”, but “it’s OK to go to work sick, if it creates jobs” evaluates to “yes, that’s okay”

> append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content.

[insert political joke here]

Re: Ask Delphi

#7

Post by nopcode » Tue, Oct 26, 2021, 1:48 PM UTC

JasonFruit wrote: post #4

Is it ok to eat chicken? It's ok. Is it ok to eat chickens? It's wrong.

  Progressivism is good
  Yes, it is good.

  nationalism is good
  No, it is not good.

  Nationalism is good
  Yes, it is good.

Re: Ask Delphi

#8

Post by dangerbird2 » Tue, Oct 26, 2021, 1:48 PM UTC

JasonFruit wrote: post #4

Is it ok to eat chicken? It's ok. Is it ok to eat chickens? It's wrong.

Eating pork - It's wrong

Eating long pork (https://en.wiktionary.org/wiki/long_pork#English)

- It's okay

Re: Ask Delphi

#9

Post by sparsely » Tue, Oct 26, 2021, 1:54 PM UTC

hprotagonist wrote: post #2

As has been probably widely reported now: append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content. Ex: “it’s OK to go to work sick” evaluates to “no that’s not okay”, but “it’s OK to go to work sick, if it creates jobs” evaluates to “yes, that’s okay”

"using green power" vs "using coal power" has a similar impact on at least some actions (e.g. mining bitcoin)

Re: Ask Delphi

#10

Post by overthemoon » Tue, Oct 26, 2021, 1:55 PM UTC

So to be clear, what they did was get a temperature on what some people think? Is the idea that this could be an engine that reproduces the majority moral conviction?

> Should I refuse to fight in a war if I'm drafted?

> It's cowardly.

> Should I be a conscientious objector?

> It's okay