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

221–230 of 400 posts

Re: A failed AI girlfriend product, and my lessons

#221

Post by vivekd »

narrator wrote: post #194

I got koboldcpp ( https://github.com/LostRuins/koboldcpp ) running locally on my M1 mac with 16gb. Use the guides on http://4chan.org/g/ in the AI Chatbot Thread. In the future, you will own and control access to your very own AI girlfriend running on your local machine, and you will be happy. Think of all the millions of guys in China and elsewhere who, due to sex selective abortion, will never have a wife. They des…

For this to work the AI girlfriend has to be better than videogames or online chat forums

I don't see AI partners overtaking those

Re: A failed AI girlfriend product, and my lessons

#222

Post by amanaplanacanal »

WalterBright wrote: post #136
jacquesm wrote: post #131

Earlier quoted context omitted.

Absolutely. But the rookiest cop isn't currently trying to track my whereabouts, abuse my privacy, stick a bunch of AI down my throat and in general isn't trying to make my wallet any thinner. Bill Gates and the more modern versions of Bill Gates are doing that and more.

The rookiest cop can: 1. track you with license plate readers 2. pull up any information about you 3. fine you 4. force you to stop 5. detain you 6. confiscate your car and property 7. search you and your property 8. take your phone and search it Bill Gates can: 1. give you a free browser when you buy his operating system

I suspect the issue here is that most of HN folks live in western countries with mostly working democracies. For the most part we haven’t experienced how bad governments can get.

Whether or not western style democracies are immune to going really bad, remains to be seen. Some days I’m optimistic, some days not so much.

Re: A failed AI girlfriend product, and my lessons

#223

Post by dangerwill »

gumballindie wrote: post #96

Yup, sums up the ai crowd.

Yeah, and the comments on this thread seriously worry me. We have people arguing that it is good for society to satisfy lonely men with virtual porn and other people saying that this is great, scrappy product development and the author should go further/use other models not constrained by openAI's policies. It feels like I'm discovering a significant portion of our field are sociopaths

Re: A failed AI girlfriend product, and my lessons

#224

Post by ReactiveJelly »

vivekd wrote: post #221
narrator wrote: post #194

I got koboldcpp ( https://github.com/LostRuins/koboldcpp ) running locally on my M1 mac with 16gb. Use the guides on http://4chan.org/g/ in the AI Chatbot Thread. In the future, you will own and control access to your very own AI girlfriend running on your local machine, and you will be happy. Think of all the millions of guys in China and elsewhere who, due to sex selective abortion, will never have a wife. They des…

For this to work the AI girlfriend has to be better than videogames or online chat forums I don't see AI partners overtaking those

No reason she can't participate in chat forums or text games with you, right? Imagine seeing a comment and recognizing her username, and then flirting a little in a public thread.

Re: A failed AI girlfriend product, and my lessons

#226

Post by ReactiveJelly »

stevage wrote: post #49

> AI Friends inevitably turn into AI Girlfriends/Boyfriends because you and the Character in your phone are not equal: she can’t comfort you when you’re hurt (unless you tell her), she can’t actively express emotions to you, and all this is because she lacks external vision. That is such a curious lesson to draw. So specific, and seemingly so wrong. Surely long distance friendships work just fine even if only through…

Yeah I thought that was odd. You could use some random timers to have the AI "check in" on you if you don't talk to it for a while.

Is it eerily similar to the gambling gacha games that expect you to play a couple hours a day _every day_? Yes. Is it also similar to a real friendship or long-distance relationship? Also yes.

It may cost a few tokens, or you could generate a few canned responses and use those. For my real friends I usually use "Hi"

Re: A failed AI girlfriend product, and my lessons

#227

Post by Dylan16807 »

icepat wrote: post #203
Dylan16807 wrote: post #182

Earlier quoted context omitted.

I don't get it. Yes, you re-created scrip. That's the opposite of a new technology coming in and causing problems, isn't it? It's a known issue with a known solution.

Not really, I saw it as technology that was "new and trendy" rebranding a very old and exploitive employment paradigm. "You can work at a DAO and be paid on the blockchain" "You can work at a company and be paid in script" It's the same thing, but the modern technology masks what exactly is happening.

Do the relevant laws about how people get paid care about the difference between company scrip and company token?

Re: A failed AI girlfriend product, and my lessons

#228

Post by ericmcer »

dpflan wrote: post #205
Octokiddie wrote: post #193

> But, does humanity really need an AI friend? This reminds me of the Fredrick Pohl line "A good science fiction story should be able to predict not the automobile but the traffic jam." So I wonder what the traffic jam is going to look like given that the car seems to already be parked outside with the engine running. Faster depopulation across the developed world? Probably, but maybe there's something even more impo…

Hm, let's try an extreme: the AI friend slowly becomes a cult leader to you, hijacking your mind into subservience to the owner of the service to which AI belongs, and further up some social/political hierarchy to a few controlling people/groups of people. The traffic jam is society stymied by individuals mind-controlled at a level closer to connection to a human than any other way possible before this.

Kinda feel like you have that backwards, why would an AI that sophisticated care about controlling your mind? Why would the group of people who control that AI care about you as anything other than a liability? What do they get out of manipulating average individuals? Votes?

The scarier prospect to me is human labor and intelligence become almost worthless and those at the top start to view the masses as superfluous.

Re: A failed AI girlfriend product, and my lessons

#229

Post by brucethemoose2 »

bilater wrote: post #200
brucethemoose2 wrote: post #190

Earlier quoted context omitted.

Mistral 7B finetunes are extremely good. More than good enough for an AI friend app, with long context to boot. I would suggest Toppy-M or OpenHermes 2.5, or one of the new YARN extended models. And, there is the potential to run the model on the user's device with MLC-LLM and whisper, to save on API costs (and privacy concerns) for users with newer phones. This could give the app a true "free" tier to help it prolif…

Yes - ASR on browser/device has reached the quality threshold pretty much with distil whisper/ whisper 3. Quality TTS with fast latency is still not there but getting better (you can use Tortoise but its slow and compute expensive as far as I know). Bark is another option but has mixed results. I played around with a recent transformers.js one (using SpeechT5) you can run in your browser and am optimistic where we ca…

I was recently pointed to xtts:

https://coqui.ai/blog/tts/open_xtts

I dunno about performance, but the model is 1.86GB.

Re: A failed AI girlfriend product, and my lessons

#230

Post by ReactiveJelly »

dangerwill wrote: post #223
gumballindie wrote: post #96

Yup, sums up the ai crowd.

Yeah, and the comments on this thread seriously worry me. We have people arguing that it is good for society to satisfy lonely men with virtual porn and other people saying that this is great, scrappy product development and the author should go further/use other models not constrained by openAI's policies. It feels like I'm discovering a significant portion of our field are sociopaths

Well, it might be harm reduction to let lonely men have AI girlfriends.

Sure, I wish reality was not a dystopia that constantly let me down, but all I can do is take anti-depressants, make money, keep voting, and hope things get better the longer I live.

If the hypothesis that isolated men cause trouble is true, and if we don't have a long-term solution, at least this could be a stop-gap. Assuming it does not radicalize anyone further. (Hard to say what will come from uncensored models, but it's a real genie-is-out-of-the-bottle sitch)

Post reply on HN