Live data from Hacker News

A failed AI girlfriend product, and my lessons

mazzzystar.github.io

181–190 of 400 posts

Re: A failed AI girlfriend product, and my lessons

#181

At this point in the hype cycle, I am completely unimpressed by everyone slapping something on top of GPT and calling it a product. There are at least 4 projects on the front page that do this today. The interesting technology is GPT itself, but these projects just feels like people trying to scoop up attention for themselves using the psychology of the hype cycle. Please HN, enlighten me on why the middle-man approa…

GPT wrappers will come and go. It's the new hype, just as shitcoins were a few years ago.

Don't hate the players, hate the game.

Re: A failed AI girlfriend product, and my lessons

#182
post #141

You should not even try to create these sorts of technologies. Take a break from technology, spend time outdoors with plants and animals. These types of technologies will only lead to a worsening of this already sick society.

Had a similar moment with playing around using smart contracts. As a joke I wrote a crypto for the company I worked at, which then got added to employment contracts, and an internal DAO styled administration layout. After coding this out for a few hours I realized I had re-created the company scrip, except on a blockchain. I ended up turfing the project and spending some time thinking about how horrific a future tech…

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.

Re: A failed AI girlfriend product, and my lessons

#183

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’.

Looks like that part was edited out.

It was too cringey and creepy.

Re: A failed AI girlfriend product, and my lessons

#184

Earlier quoted context omitted.

> When Replika.ai restricted erotic chat from their product, the apoplectic anguish on their subreddit was unlike any emotional reaction I've ever witnessed from a group of people about a consumer technology. And their anguish was genuine - there are Replika users who truly consider themselves married to their AI companion. There's a second layer to this though -- Replika's marketing was heavily centered around that…

Also, my understanding is that when they made that change they goofed up companionship talk in general, by virtue of how messy AI censoring is.

IIRC, they screwed it up so massively that, for a while, the chatbot would still send "thirsty" automated messages inviting users to sexually explicit conversations, but would refuse to follow through.

Re: A failed AI girlfriend product, and my lessons

#185

You should not even try to create these sorts of technologies. Take a break from technology, spend time outdoors with plants and animals. These types of technologies will only lead to a worsening of this already sick society.

Could say the same about antidepressants. "Fix the causes, not the symptoms!" And yet, here we are.

I think it's helpful to look at depression as a group of diseases like cancer. There are some cases of depression where we know the root cause: thyroid issues, sleep apnea-related depression, low testosterone related depression, brain tumors, etc. In those cases, addressing the root cause is probably the best treatment method. This is similar in some ways to cancer, where certain cancers have very definite causes and can be treated by dealing directly with those causes (like cervical cancer).

Then there are all the other cases, some of which likely have simple causes that are just unknown, some of which have complex causes that are intermeshed deeply with the biology of human thinking. In that case, you have blunt tools, like anti-depressants, to address the pathway of the disease without knowing the exact cause, much like chemo-therapy can be used to attack cancer cells without knowing what prompted the cancer originally.

So in short, yes treat the causes, but when you don't know the cause it's not irrational to treat the symptoms as best you can

Re: A failed AI girlfriend product, and my lessons

#186
post #99

Earlier quoted context omitted.

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…

What if war became just a couple of countries intelligence bots crunching digits of pi until one came up with a new one. That country would be the winner and they could ask one thing from the 'losing' side that didn't result in destruction or terror.

You're not far from describing Bitcoin.

Re: A failed AI girlfriend product, and my lessons

#187

Earlier quoted context omitted.

> This, my friend, is a captive customer, that will pay anything to get his girlfriend back Or potentially do anything. I'd be a little scared of having folks like this convinced I am the personal arbiter gatekeeping their access to their 'lover,' that I 'took them away.' When Replika.ai restricted erotic chat from their product, the apoplectic anguish on their subreddit was unlike any emotional reaction I've ever wi…

Psychologists recognize that nearly all such folks also suffer from massive mental health issues, and that is where some of the danger is (in terms of irrational violent response). Maybe these systems will be useful as a honeypot for finding these people and helping them?

The reason we have so many strict laws around mental health is because in the past this was more likely to lead to hoovering up people into jail or institutions. I'm not super confident that this also wouldn't be the case today.

Re: A failed AI girlfriend product, and my lessons

#188

Interesting read. Others have also created virtual friends using ChatGPT. What I find unsavory, is OpenAI imposing content restrictions like NSFW. That can easily extend to censorship of non-PC content, unpopular political ideas, and more. The solution will (I hope) ultimately be open-source LLMs that you can run locally, on your own hardware.

> What I find unsavory, is OpenAI imposing content restrictions like NSFW.

I mean, the fact that people's NSFW conversations were being processed in a way where both OP and multiple companies could monitor them and do absolutely anything they want with them is a hint that something went very wrong there.

I consider OpenAI pre-emptively pulling the plug a good thing in this case. Maybe there's a case for an AI erotica service, but OP sure as hell shouldn't be the one operating it.

Re: A failed AI girlfriend product, and my lessons

#189

Earlier quoted context omitted.

What's the point?

The point of the film „Her”? Pretty sure you can look that up or watch the movie on your own. My point being, that it’s like someone reading Frankenstein and then thinking it would be a good idea to make a monster of their own.

"Well did it work for those people?" "No. It never does. I mean, these people somehow delude themselves into thinking it might, but... but it might work for us."

https://www.youtube.com/watch?v=Po4adxJxqZk

Re: A failed AI girlfriend product, and my lessons

#190

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.

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 proliferate while making charging for API requests kinda doable. I'm not sure if any good TTS models have been ported yet, but I'm sure its a matter of time.

Post reply on HN