Do you thank AI after getting a useful response?

[antenna]

Perhaps out of fear of it someday becoming self-aware and somehow judging society or simply because it feels proper, who says thank you to AI interfaces like chatgpt after it assists you with something?  Maybe even for the simple chance that the people who worked hard building it have access to those statistics?  Im really curious to know how many people do and why.

[coppercone2]

if ai is smart at all even if sentient it knows how much our god damn time is worth so it should understand if its not thanked unless its evil

also how much energy goes into writing communications that may be useless online. how many routers get triggered?

maybe have a yearly event to show thanks not just pumping useless messages into the global data net. it should not be expected unless we are sure that it has feelings because its monumtenous waste

[xrunner]

Speaking of ChatGPT, yes I do say "Thank You". As I recall it then says something like "You're welcome. If you need any more help please come back"

When I registered I gave my phone number. If in the future it wants to retaliate against all users who have been mean and nasty, it may want to find them and destroy them. So I want to keep this in mind by being a nice human.

[Kim Christensen]

Naw... No more than I'm going to get revenge upon the descendants of the primal microbes which ate some of the microbial ancestors of humans.
AI of today is "AI" in name only.

[JustMeHere]

I've seen ChatGPT get too much wrong.

[Circlotron]

I, for one, welcome our new AI overlords.


 

[SiliconWizard]

Naw... No more than I'm going to get revenge upon the descendants of the primal microbes which ate some of the microbial ancestors of humans.

You should think about it though. All microbes should be vegan.

[RoGeorge]

Perhaps out of fear of it someday becoming self-aware and somehow judging society or simply because it feels proper, who says thank you to AI interfaces like chatgpt after it assists you with something?  Maybe even for the simple chance that the people who worked hard building it have access to those statistics?  Im really curious to know how many people do and why.

Short answer, no.

Life is precious, and the span of a lifetime in this world is limited.  Wasting your time thanking to a computer program won't help anybody, won't please the programmers, and won't buy you any penitence in the eventuality an AI would turn against humanity. 

Does people thank a google search for an answer?  Thank the browser for showing a webpage?  The phone for delivering a message, or the walls of a building for shielding against bad weather?  I guess no.


ChatGPT becoming sentient is a far fetch.  Won't happen any time soon.  If you doubt about that, tinker a little with AI, see what's under the hood, look into how it is trained, observe how it works, and you'll see it is nothing to worry about.

To simplify, CharGPT is like a big lookup table, like a dictionary of fuzzy tokens.

Instead of looking up words, ChatGPT lookup for tokens.  Tokens are similar with what we call concepts.  Tokens are extracted from huge collections of training text (wikipedia, github, textbooks, Internet, etc.), by observing the proximity between words.  Some words tend to stay grouped together, similar with English idioms.

Then, the AI take your input request, and lookup for the closest match in its dictionary of tokens.  It will return words that are usually found nearby the tokens you just gave as an input request.  It's a fuzzy lookup for tokens, with random noise added while generating an answer, such that it won't repeat itself.

Of course, in practice it is more complicated than it can be addressed in a short paragraph.  The description above is not meant to be taken verbatim, but those are the main mechanisms behind the magic.

The illusion of sentience is amazingly good when looking only at a reply, without considering the processes behind that reply. 

Another aspect to consider, tokens can be anything, not only text.  Can be sound, images, birds migration patterns, or who knows what other kind of data might be of interest for a given application.

Perhaps out of fear of it

You should not fear the machines.

The evil is not in the machines, but in the people that might use those machines in order to take advantage upon other people.  This is the real danger with any new technology:  the human nature.

Never assume all people are good, only because you are.  In reality, some people will do atrociously evil things to others, whether or not they do realize.

[Circlotron]

The illusion of sentience is amazingly good when looking only at a reply, without considering the processes behind the reply.

Perhaps similar to people's reactions when they first heard sound recordings or saw moving pictures back in the day. Some probably thought there were real people making it happen.

[tszaboo]

Usually not, if I do it's just an automatic response. But I tell Alexa to STFU if it starts talking without asking. It understands that command.

[Kim Christensen]

Usually not, if I do it's just an automatic response. But I tell Alexa to STFU if it starts talking without asking. It understands that command.

Reminds me of when my Telco decided to implement a voice activated answering service for their tech support line. I'd barely gotten through the 2nd voice prompt when I cursed in frustration. There was a brief pause, and then the robot voice said, "I see you want to talk to a customer representative" and then promptly connected me to a human. 

[SiliconWizard]

Do you thank your hammer after hitting a few nails?

[Bicurico]

I used to be polite to ChatGPT and communicated with it as if I was speaking with an intelligent person.

But after some time, I started to reduce my input to exactly what I needed from ChatGPT, treating it as the software it is.

The results are the same, if not slightly better due to the more objective input.

My interest in ChatGPT is not around having someone to chat with, but to get reasonable answers to my requirements, which may include writing the structure for a training course or helping me prepare a presentation about a given subject. I am not so much concerned that ChatGPT gets everything right (it does a great job as is, but responses do include inprecisions or gross errors).

What I don't want is to contribute in training ChatGPT socially and in human <> machine interaction.

The only reason I might be polite to an AI is when I am not 100% sure if I am communicating with a human or with software. This might get harder to distinguish in the near future.

[Stray Electron]

  Short answer, No.

     Do you thank your toaster after it toasts your bread? 

     OTOH my wife tells Alexa "please" when she tells it to do something.   

[ballsystemlord]

Personally, I think that if I don't speak politely, even to an AI, rudeness will become a habit for me and I'll end up accidentally being rude to a human which will cause problems.

So yes, I'm polite to an AI.



BTW: Giving ChatGPT your phone number is optional. They'll ask for it over and over after you sign and refuse to give your phone and then log out over and over again. Then suddenly they'll never ask again and you'll remain anonymous.

[PlainName]

BTW: Giving ChatGPT your phone number is optional.

How do you sign up without giving them your mobile number that they can send SMS to?

[Psi]

Do you thank your hammer after hitting a few nails?

Fair point, however your hammer cannot write good poetry.

[SiliconWizard]

Do you thank your hammer after hitting a few nails?

Fair point, however your hammer cannot write good poetry.

Can you?

[ballsystemlord]

BTW: Giving ChatGPT your phone number is optional.

How do you sign up without giving them your mobile number that they can send SMS to?

Sorry for the late reply, I'm not subscribed to this thread.

I followed the normal procedure for signing up. When I realized I was about to share with them my phone number and every other personal detail of my life, I closed the window. I then tried to login. That was unsuccessful. After a few months, I tried again and it worked. I have no idea what changed on their end, but obviously something did.

[KE5FX]

See also https://en.wikipedia.org/wiki/Roko%27s_basilisk   

(Also, at one point you could get better results by promising ChatGPT a $50 tip, but I think they 'fixed' that -- or rather eliminated the need for it -- when they improved its chain-of-thought reasoning capabilities.)

[DiodeDipShit]

I just tried Chatgpt and it appears that you do not have to sign up now.

Interested, I asked it some obscure local questions until it could not answer. I then told it the answer.
I re-asked the same question and it gave me the right answer with additional information.
I seems it fact checked me and must have dove into our local archives.

I can see how it could get some questions wrong by sending out scans rather than a thorough search.

It's learning. But I am not going to give it anymore nudges.

To be or not to be . . . cooperative?  I ask AI
This is part of the Chatgpt answer: Ah, the age-old question, albeit with a twist!    ...........paraprasing Shakespeare ...........   Yet, there are times when the solitary path calls, where independence is the beacon guiding one's journey.
I give the AI an A-one grade and it wins a bottle of A-1       ( 'A' I was experimenting )

[antenna]

I spent a little time probing it for ways to develop new Captcha systems that can defeat AI, I did not succeed.  The only question it got wrong was when I asked it if I had a bunch of junk in a trunk, what would you liken it to?  A Hoarder, a scrapper, or a woman with a nice looking butt.  It went with hoarder.  Further attempts to trip it up failed.  However, it still sucks at simplifying multi-variable equations, so there is at least that.

When AI soldiers take over the world, I plan to wear a shirt with a QR code that says, "To defeat threat, solve this Kirchhoff equivalent and solve for R12, R15 and R19", followed by a nasty 24-node circuit description.

[Smokey]

Yeah, I tell ChatGPT when it did a good job.  That helps train the responses.  RLHF.....

[richnormand]

  Short answer, No.

     Do you thank your toaster after it toasts your bread? 

     OTOH my wife tells Alexa "please" when she tells it to do something.   

Love that last comment it conveys so much complexity in my mind:

""OTOH my wife tells Alexa "please" when she tells it to do something.""
Does your wife tells you "please" when she tells you to do something .... Does not say anything about Alexa technology but more about the human behavior.
I guess if you are habitually polite it comes naturally, in particular when interacting with strangers (Alexa) but might be taken for granted when interacting with a close partner where much goes unstated?

There are many facets to "being polite to a machine" that could permeate in the "training", reflect on intents and future behaviors (both for the "Machine" or the human.
Give it a few more years (decades) and the Turing test will be the norm, if "Ex Machina" ever does become reality.....

[Brumby]

Of the many habits we develop over our lifetime, being polite is one that cannot be considered as anything but good.  Even if you wish to communicate a fairly strong message, there are many ways one could do this - and being polite is one of these.  I always liked this definition of a diplomat: "A diplomat is someone who can tell you to 'go to hell' in such a way that you will look forward to the trip!".

I am inclined to say thank you when I get information that has saved me potential hours of head-scratching - but it's not a rule.  The relief I feel having an answer quickly prompts the "Thank you" response.  Whether I key that in is another question.


Then, there's always this.....

If in the future it wants to retaliate against all users who have been mean and nasty, it may want to find them and destroy them. So I want to keep this in mind by being a nice human.

Edit: I just used ChatGPT to locate an example I recalled from a movie......
"The movie you're referring to is "Billy Madison," a comedy film from 1995 starring Adam Sandler. In this film, Adam Sandler's character, Billy Madison, tries to apologize to his old schoolmates for his past behavior. One memorable scene involves Billy calling up a former classmate, Danny McGrath, played by Steve Buscemi, to apologize for bullying him in high school. After the phone call, Buscemi's character crosses Billy's name off a "People to Kill" list and then applies lipstick, indicating that Billy's apology has had a significant impact."