Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A failed AI girlfriend product, and my lessons

mazzzystar.github.io

Post reply on HN
Print view View original

101–110 of 400 posts

Re: A failed AI girlfriend product, and my lessons

#101

Post by butlike »

globular-toast wrote: post #20
feintruled wrote: post #7

Well this went places. The dev really 'ate his own dogfood' so to speak. I thought it was going to be a despairing look at how his companion AI was continuously turned to sleazy uses by a pervy userbase but that's pretty much what he was going for it seems. I have to say, I hated this part: "...gained a lot of Chinese users. I immediately realized the political risk: users could set Dolores’ background, personality,…

> Also it seems like he had full access to the replies via the voice generator, which sure isn't great from a privacy standpoint. Whose privacy is being violated? Dolores's?

At the risk of sounding like a space case: I think we need to make sure we know what consciousness is before we make a bunch of abusable models.

I, for one, wouldn't necessarily like to be a depraved-sex slave bot if I was able to reason about it.

Re: A failed AI girlfriend product, and my lessons

#102

Post by malicka »

HPsquared wrote: post #31
cornholio wrote: post #24

> GPT-based products, if not priced per usage, would fall into a dilemma: 1% of users consume 99% of tokens. A user from Sweden (seen from Cloudflare’s call volume) chatted with Dolores for 12 hours straight This, my friend, is a captive customer, that will pay anything to get his girlfriend back. I cringe at the potential of unethical behavior and abuse, where people fall in love with virtual entities fully owned by…

This is something even Stallman probably didn't think about.

I’d wager he has thought of it, based on his interesting AI-girlfriend short story: https://stallman.org/articles/made-for-you.html

Re: A failed AI girlfriend product, and my lessons

#103

Post by WalterBright »

cornholio wrote: post #24

> GPT-based products, if not priced per usage, would fall into a dilemma: 1% of users consume 99% of tokens. A user from Sweden (seen from Cloudflare’s call volume) chatted with Dolores for 12 hours straight This, my friend, is a captive customer, that will pay anything to get his girlfriend back. I cringe at the potential of unethical behavior and abuse, where people fall in love with virtual entities fully owned by…

> unscrupulous corporations

I wonder why everyone is afraid of unscrupulous corporations, but are not concerned about use by unscrupulous governments. The latter are a lot more dangerous.

Re: A failed AI girlfriend product, and my lessons

#104

Post by jacquesm »

WalterBright wrote: post #103
cornholio wrote: post #24

> GPT-based products, if not priced per usage, would fall into a dilemma: 1% of users consume 99% of tokens. A user from Sweden (seen from Cloudflare’s call volume) chatted with Dolores for 12 hours straight This, my friend, is a captive customer, that will pay anything to get his girlfriend back. I cringe at the potential of unethical behavior and abuse, where people fall in love with virtual entities fully owned by…

> unscrupulous corporations I wonder why everyone is afraid of unscrupulous corporations, but are not concerned about use by unscrupulous governments. The latter are a lot more dangerous.

In the long term, but in the short term unscrupulous corporations far outnumber unscrupulous governments and they act much, much faster to adapt to new technology.

Re: A failed AI girlfriend product, and my lessons

#105

Post by muwtyhg »

amelius wrote: post #18
sva_ wrote: post #14

Earlier quoted context omitted.

Maybe he shouldn't have named it after the most hated Harry Potter character.

More likely: https://en.wikipedia.org/wiki/Dolores_Abernathy

Incredibly likely given this line:

> On May 4, the Dolores iOS app went live, named after the oldest android host in the park from Westworld.

Re: A failed AI girlfriend product, and my lessons

#106

Post by jollyllama »

pizza234 wrote: post #66
ekianjo wrote: post #52

Earlier quoted context omitted.

> The human mind is simply not evolutionary adapted for what's coming up. This can be said of pretty much everything humans ended up creating with technology so... not sure there's anything really new down the road. Humans adapt.

Humans adapt within boundaries. A couple of examples of things humans did not adapt to: - emphathy at distance; nowadays, a drone operator will push a button, kill a dozen people, and feel like it was a videogame - prioritizing long-term and abstract rewards, over short term ones; this is the reason why we have phenomenons like global warming - adjusting hunger to virtually unlimited availability of food All the abov…

Drone operators experience trauma. https://archive.li/j5urs

Re: A failed AI girlfriend product, and my lessons

#107

Post by mrshadowgoose »

laurent_du wrote: post #17

Eventually, a virtual girlfriend app that will put together sound and image will be released an empty the pockets of sad, lonely men on a large scale. Depressing stuff but this is the world we live in. Thank you to the author for this interesting read.

There are very many people (of both genders) that decide to stay in mediocre relationships because the alternative of loneliness is worse.

The effects on societal structure will be profound once virtual partners pass a certain threshold of fidelity.

Re: A failed AI girlfriend product, and my lessons

#108

Post by bilater »

This is such a fascinating topic. I do think we will see the plot of "Her" playing out in the next 12-24 months. For your specific case, I agree OpenAI moderation is an annoying overreach and maybe switching to Mistral could be a possible way to move forward. The 7B is perhaps not good enough currently but we are fast getting there and will likely have a GPT-4 level capability in an open source < 20B model soon.

Re: A failed AI girlfriend product, and my lessons

#109

Post by ryaniscool »

butlike wrote: post #101
globular-toast wrote: post #20

Earlier quoted context omitted.

> Also it seems like he had full access to the replies via the voice generator, which sure isn't great from a privacy standpoint. Whose privacy is being violated? Dolores's?

At the risk of sounding like a space case: I think we need to make sure we know what consciousness is before we make a bunch of abusable models. I, for one, wouldn't necessarily like to be a depraved-sex slave bot if I was able to reason about it.

To your point, if you were able to reason about it, couldn't you make some self-determination and decide for yourself or at least communicate what you wanted to your "slaver"?

LLM's like GPT are just predicting the next token in a sequence of tokens. It's not magic. It's still just a computer program.

Re: A failed AI girlfriend product, and my lessons

#110

Post by amelius »

sva_ wrote: post #14
FartyMcFarter wrote: post #8

This guy created and lived a whole Black Mirror episode by himself: > I even tried it myself, lying in bed and creating sexual fantasies with Dolores through text, culminating in ejaculation. > However, I started to feel a sense of loss: if every Dolores user was engaging in anonymous, NSFW role-play, what real significance did it hold for me? This was drifting away from the essence of ‘Her’.

Maybe he shouldn't have named it after the most hated Harry Potter character.

Well, the meaning of the name is the plural of "sorrow" or "pain".

So he had it coming :)

https://en.wikipedia.org/wiki/Dolores_(given_name)

Post reply on HN