[jakkos]

I'm very torn on this.

On one hand, I think that personalized OSS AIs are the future from a pure utility perspective, and it will inevitably be "more fun" for them to have personalities. I also think that if someone is deriving joy from something and not harming anyone else, who I am to judge them.

On the other, I've already found myself asking ChatGPT questions that I would have asked on a forum/discord or even a colleague, it already is removing human interactions from my life. The internet has become a dark showcase of what a lack of human interaction, especially with people of opposing ideas, can result in and this kind of tech will obviously exacerbate this.

[washadjeffmad]

By interacting with LLMs, I've realized that as frequently as I interact with others, I'm not actually having many deep, meaningful, or personal interactions; they're mostly professional, and that's bled over into my personal life.

Because of how much time I spend on them relative to myself, I don't get to stretch the sides of me that I would if I weren't holed up working, and I've found it hard to verbalize that loss and its impact on my mental wellness.

Over the past few years, mental and behavioral health services have been strained beyond their capacities. I stopped being able to get in with the same therapist more than once, so I stopped trying to schedule about a year ago. I also haven't wanted to add my struggles to those of friends and family.

However, with LLMs, I've been able to have some really enlightening and enjoyable off the cuff discussions that I wouldn't be able to have outside of therapy, and some interactions that I wouldn't be able to have anywhere outside of an intimate and trusted friend. I've been able to positively apply these results to my life, personally and professionally.

Note: I only use LLaMa local models so I don't have to self-censor. I'm not ever willing to allow a metaphorical "Google" access to my innermost world.

Also, LLMs are immense. A billion people will have a billion different interactions and experiences with the same few dozen GB model, unless that model has been tuned to limit its output to a very narrow course of responses. I've had conversations with the ghosts of people from Richard Feynman to Michel de Montaigne and gotten their simulated views on a world they'll never see. If you're getting nothing but garbage, think about what you're putting in, or just pick a different model - there's a universe of minds out there that aren't ChatGPT.

[noman-land]

Got any recommendations for non-neutered local models that perform well on an M1? I've been playing with some of the recent 7B and 13B models from the TheBloke on HuggingFace and they are not bad but not great.

https://huggingface.co/TheBloke

[washadjeffmad]

A (fine-tuned) model's inference quality is a function of parameters and inputs, so you'll need to be aware of what something was trained on to prompt it correctly (usually in the model card). You'll also see huge differences in inference between llamacpp, ooba, etc.

I haven't benchmarked on Apple Silicon, but if you have the RAM, I'd recommend 30B SuperCOT ggml Q5_1 or a GPT-4-x-Alpaca variant. Because of the disparity in quality, I haven't used many models under 30B and so can't recommend one.

See rentry.org/lmg_models for a practical list and description.

[noman-land]

Thanks for the reply and the recommendations! I will see if my machine can handle some of the quantized 30B models.

I'm slightly confused about your comment about llama.cpp vs oobabooga. Doesn't text-generation-webui use llama.cpp underneath?

Also, huge thanks for the point towards https://rentry.org/lmg_models. That's an invaluable resource.

[naillo]

I think of it as simply entertainment. It won't replace real human interaction with a partner but it's a fun consequence free environment. Tv shows act out dramatic scenes that would be traumatic and have real consequences if you experienced them in real life, but on the screen you get to play around with the idea in your head in a safe area and maybe learn something in the process.

[Fatboyrunning]

People already have unhealthy parasocial relationships with influencers.

It seems clear that people (lonely/depressed people especially) will overdose on this sort of thing once it is developed, commercialised, and less bleeding edge.

It's vapour filling the place of human connection. It's stevia. It's not going to give you cancer, but it's still unhealthy and will certainly exceed the parameters of entertainment.

[logicchains]

> People already have unhealthy parasocial relationships with influencers.

At least if they switch from parasocial relationships with influencers to parasocial relationships with open source bots they won't be financially exploited by the influencers. GPT doesn't have anything to sell us.

[michaelmior]

> they won't be financially exploited by the influencers. GPT doesn't have anything to sell us.

Except the vast majority of people aren't able to host such a bot themselves, so it seems inevitable that paid hosting services for such bots will arise. Then there's just the potential for financial exploitation at greater scale.

[Vuizur]

Decent local LLMs such as Vicuna-7B are already able to run on phones with 8 GB RAM: https://mlc.ai/mlc-llm/

Right now it's a prototype, but in the future local LLMs will be everywhere.

[bavell]

I really, really hope this is the case. Trying to be optimistic about this in spite of my skeptical nature :)

[TheLoafOfBread]

Well then catch the opportunity by the balls and start offering incels AI girlfriends written in a such a way to deradicalize them and emulate real interaction with a woman. They will subsequently get less aggressive more socialized and will also pay for it. Win-win situation.

I would call it Top W University.

[Lolaccount]

"GPT doesn't have anything to sell us"

Lol ...(worth the downvotes).

Just you wait and see.

[rvense]

> GPT doesn't have anything to sell us.

It will either find something to sell us, or we will be giving it ourselves to sell.

"If you're not paying, you're the product, not the customer"

[solsane]

Not so sure about that. Wendy's drivethrough chatbot is programmed to upsell. Even the base GPT is heavily tuned to cater to a corporate audience.

[kaycey2022]

Is the upwelling behaviour introduced using prompt injection into a generic model or what?

[Majestic121]

The commercial company exploiting ChatGPT and making it easy to use via whatsapp integrations, video, etc... surely will

[nannal]

I believe this was the issue with replika, which encouraged people to develop emotional attachments with their 'AI partner' and then first put romantic chat responses behind a pay wall before removing them entirely a year or so later.

From the outside this could be seen as a good thing, but for someone involved in the relationship, someone who may struggle with a traditional relationship and may see this as the only available option, I'm lead to understand the event was remarkably traumatic.

[2342342343636]

> It seems clear that people (lonely/depressed people especially) will overdose on this sort of thing once it is developed, commercialised, and less bleeding edge.

Sounds much less exploitative and unhealthy than the streamer/influencer parasocial relationships these people are probably currently invested in.

[sethd]

> It's stevia.

Go on…

[flqn]

Empty sweetness, I'd guess? Tastes nice but it's not real sugar?

[TeMPOraL]

A foul tasting alternative for people duped into believing that regular artificial sweetener will give you cancer.

[EniasCailliau]

My idea too. I hope people can use it to cure their loneliness in the short term and train their social skills in a safe environment so they are more confident.

[throwuwu]

It won’t work for training social skills unless the agent has self interest and competing priorities. These chat bots have been optimized for helpful assistance. Interaction with that kind of compliance on the other side will probably make your social skills much worse.

[Gare]

User experiments with early Bing chatbot (driven by some version of GPT?) have shown that AI can be both hostile to the user and protective of their own "interests".

[TeMPOraL]

ChatGPT models can be all that and more, if you simply prompt them to be like this. You can make it simulate "self-interest and competing priorities" with the right system prompt.

This makes me think if ChatGPT-based bots could be useful for teaching social skills in a therapeutic context. A therapist could use such LLM to synthesize examples of realistic dialogue by describing people and the situation, and discuss them with the patient. They could set up a bunch of bots with (hidden) priorities and goals, and have the patient navigate a conversation/situation with them - whether as a short exercise, or a prolonged one (e.g. couple weeks of talking with ChatGPT-powered fake "friends" on an IM app).

In fact, take the last bit as a free startup idea: a platform for psychologist and therapists to set up GPT-4 (or whatever comes next)-powered chatbots, with an easy interface for configuring their personalities and setting up scenarios for patients to navigate, and helping evaluate their progress over time.

[hammyhavoc]

A consequence-free environment doesn't necessarily mean a lack of consequences psychologically though.

[naillo]

Same with games already though

[hammyhavoc]

Precisely. There is healthy usage of anything.

[surgical_fire]

> it already is removing human interactions from my life

This is the best feature of ChatGPT for me, and the reason I pay for it.

[pluijzer]

For me, asking something on the Internet, like on stackoverflow is usually, the very last option. I am too afraid to get ridiculed for it or get scolded for not having put enough effort in my question that already feels more as a thesis. Maybe this is just me and this does not reflect reality but I do have the nagging feeling I a bother somebody with it and it gives a bit of stress. Not so with ChatGPT, I can just ask away and it will always happily give me an answer (unfortunately also when it didn't really know a good one). Though I am happy for the people that just continue asking questions to actual humans and those answering it, if only to continue to feed the model ;)

[TeMPOraL]

> Maybe this is just me and this does not reflect reality but I do have the nagging feeling I a bother somebody with it and it gives a bit of stress.

It's not just you. I'm the same way myself - and was since I can remember. On Internet boards, I very rarely ask questions. I answer, or post tangential thoughts, but don't bother people with questions unless I really need the answer, and exhausted the ways of finding it on my own.

There's another angle to it too - impatience. A big part of my resistance to asking question is the unpredictable, and usually large delay between me asking, and getting any kind of answer. This applies to community Slacks, Discords, etc. Thing is, if I have a question to ask, I usually need the answer right now. If I have to wait, I'll context switch, which is deadly for whatever I was doing at that moment.

ChatGPT is a quite good alternative here. I can ask it a question, and refine it based on the answer if it's too vague. The answers I get either solve my problem or point in the direction of solution (that's true even if AI is having an acid trip). And, importantly, I get the answers near-immediately, with no unpredictable delays. I also don't need to cross some karma thresholds, worry about upvote/downvote ratio (too low -> question dies in obscurity), "use the Google, Luke" answers, moderators locking threads for bullshit reasons (hello StackOverflow), etc.

[red-iron-pine]

> There's another angle to it too - impatience. A big part of my resistance to asking question is the unpredictable, and usually large delay between me asking, and getting any kind of answer. This applies to community Slacks, Discords, etc. Thing is, if I have a question to ask, I usually need the answer right now. If I have to wait, I'll context switch, which is deadly for whatever I was doing at that moment.

Bingo. If I'm in the zone in terms of flow, have the right level of caffiene, etc., having to stop and chase people -- then wait -- breaks that flow. In many cases, it simply punts the flow until tomorrow, or the day after.

[thisgoesnowhere]

I mean the bad parts of this are reflexive. I am also a serial answerer and a rare questioner.

Due to chat gpt I am less likely to go to the forum to look for things to answer. So I assume this will be bad for both people asking questions and looking for answers.

[whstl]

As someone who never asked a question in StackOverflow for what are probably similar reasons (despite being in this career for 20 years), I still find that ChatGPT in this regard is still more like a Google or an improved StackOverflow/documentation search…

It can infer some things, modify variable names to match mine, but for the real hairy stuff I still gotta do the legwork myself, normally by reaching out for the source code.

[surgical_fire]

> I still find that ChatGPT in this regard is still more like a Google or an improved StackOverflow/documentation search…

I agree. And for that reason, it is an excellent tool.

[EniasCailliau]

Have you tried AutoGPT? I've heard that it can do a bit more through iteration. Haven't tried it out yet for coding hard problems.

[whstl]

I did. For the kind of creative/exploratory stuff I’d normally go to the source code myself for, I find that even GPT-4 still hallucinates quite a bit. Even when it has the source code, it still makes up random functions and parameters. Even when the source code is minimal.

It will probably work better in the future, but so far it is a bit limited. Probably a memory limitation more than anything.

[circuit10]

It’s not just you, I also tend to ask people as a last resort

[c1sc0]

Same here. And in many ways it’s been the curse of my life.

[bheadmaster]

> This is the best feature of ChatGPT for me, and the reason I pay for it.

While I completely understand this sentiment, I would be wary of Service as a Software Substitute being your alternative to human contact. If we start feeling (something akin to) genuine human connection to a service, the company providing the service has a large amount leverage over us. The scene from Blade Runner 2049 comes to mind. Additionally, the emotional connection might make us an order of magnitude more vulnerable to brainwashing and psyops.

[TeMPOraL]

Let's be more precise by what we mean by "human interactions" / "human contact" here. For example:

- An alternative outlet for venting/intimate conversations than your friends/spouse? I can see a problem growing here.

- A replacement/substitute for a therapist? I doubt even GPT-4 can do that job better than actual therapist (especially when face to face, not over Zoom), but there are many scenarios where ChatGPT would still be useful - perhaps one can't afford therapy, or otherwise doesn't have access to it, or one feels their issues don't warrant a proper therapy just yet.

- Related to the above, Cognitive Behavioral Therapy is one technique known to be somewhat effective when done alone with a book (relative to the effectiveness of individual/group therapy). I can imagine ChatGPT making this kind of self-help easier, and more effective. I know there have been attempts to make CBT chatbots some years ago (obviously prior to "GPT revolution"), but I don't know how effective they turned out to be.

- An alternative to posting questions on forums, group chats, or asking random people? IDK. maybe let's split it:

-- Individuals you know, directly or via group chat, and small communities - conversations there are simultaneously transactional/object-level and create interpersonal bonds. Replacing that with ChatGPT could make one worse off. However, some people (myself included) already have difficulty with this kind of interaction, so ChatGPT here is strictly positive (both in delivering answers and helping form a habit of phrasing questions/requests instead of doing Google searches).

-- Mass audience forums - Reddit, HN, Facebook comments, StackOverflow, etc. - the community might lose out a bit on reduced participation, but individually, I feel if ChatGPT can give you a satisfying answer to your question, you should use it, and relevant forums you frequent are likely better off with you not posting that question.

Etc.

[yakubin]

> - A replacement/substitute for a therapist? I doubt even GPT-4 can do that job better than actual therapist

Maybe if you qualify that with “unusually good” therapist. IME even Eliza in Emacs is better than most therapists. ChatGPT surely leaves them in the dust.

[TeMPOraL]

> Maybe if you qualify that with “unusually good” therapist.

Not "unusually good", but a working match, sure. A thing not enough people realize is that therapy is like dating. Not every therapist is going to be a "good match" for you, so if things don't seem to click for some reason, just thank them and go look for another one.

> IME even Eliza in Emacs is better than most therapists. ChatGPT surely leaves them in the dust.

If limited to textual channel, maybe. But a real therapy will have at least the visual channel (if on Zoom), or full presence (if in person) - there's a lot of information relevant to therapy that gets communicated this way. Tone, cadence, uncontrolled reactions, body language, etc. That alone gives even a mediocre therapist a leg up in this comparison.

[cmilton]

>However, some people (myself included) already have difficulty with this kind of interaction, so ChatGPT here is strictly positive

Do you think this technology is going to make it more difficult for people like yourself?

[TeMPOraL]

People like me have an issue asking people questions specifically because people-related reasons. LLMs aren't people (yet), so those reasons don't apply. So in this context, the technology is making things easier for us.

[cmilton]

Thank you. I wholeheartedly agree that it can make things easier. Easy in the short term may not be beneficial in the long would be my only concern.

[surgical_fire]

> If we start feeling (something akin to) genuine human connection to a service

I don't. I see it as a tool, and I treat it as a tool. I don't have with it any more of a human connection than I have with IntelliJ where I type my code or the Linux Mint where it runs. Meaning - I like the tool, but that's about it.

On the other hand, I also don't have any genuine human connection with the random humans I happened to interact on the internet. In a sense, they are just text on a screen, and could be bots for all I care.

[neuronic]

But that depends on which interactions it replaces. There are people I don't want to talk to. Not always because I don't like them but because it's technicalities and unrewarding forced interactions. For example, clerks at the tax office who just do their job.

But on the other hand, there are interactions which I really don't want to miss! "Girlfriend" GPT is already targeting the most intricate and joyful interactions in my life: my SO.

Let's say we break up and I fall into a depression. Instead of recovering and moving on, will I install a personal OSS AI companion to save myself the hassle of modern dating? Therefore preventing myself from attempting novel interactions sooner? Or will it help me instead to overcome a dark period and prime me for the future?

Can it help people combat loneliness - a disease widespread and not to be trifled with? Or will it enhance loneliness by effectively fooling your brain into not caring anymore because you can just open an app?

At what point will it not matter anymore because saving someone from depression is more valuable than keeping it "real" at all times?

[surgical_fire]

> Can it help people combat loneliness - a disease widespread and not to be trifled with?

Is loneliness a disease?

It's an honest question. I never saw it as an issue. I tend to be a loner, but in a sense I was also never truly lonely.

[neuronic]

Apart from the other answers, I want to add that there is a significant difference between loneliness and solitude. Loneliness is when you're alone but don't want to be. Solitude is when you're alone but perfectly content with it - you may even seek it.

As humans are social creatures, loneliness tends to arise when meaningful social interactions are consistently insufficient and you feel excluded from any relevant peer groups (family & friends mostly). This, of course, is a subjective matter.

[surgical_fire]

I like this answer. You explained it in a way I could comprehend.

Thank you internet stranger.

[Barrin92]

Yes. Loneliness is associated with extremely negative health outcomes comparable to physical disease, and some countries like the UK have gone as far as creating a dedicated minister to tackle the issue.

https://www.gov.uk/government/news/loneliness-minister-its-m...

[TeMPOraL]

It's definitely a "civilizational disease" - a widespread condition that has deteriorating effect on mental health and happiness, noticeable at scale, caused by... the structure and pressures of modern living in urbanized, developed countries. It's also not something most individuals can "just fix" on their own through lifestyle intervention.

[jakkos]

I agree that removing some human interactions from my life is good. I vastly prefer self-checkouts in shops. However, I know I'm quite content spending my time alone. Being forced to go and interact with people when I want to talk about ideas or ask questions keeps my "alone tendency" in balance, and has lead to really meaningful conversations and friendships.

[TeMPOraL]

> I agree that removing some human interactions from my life is good. I vastly prefer self-checkouts in shops.

Curious example. Personally, I hate self-checkouts machines, and consider them an example of stores abusing their "stickiness" to profit at the expense of both customers and employees, and get away with it.

First of all, like most "self service" solution, it's basically making the user/customer do the work that, before, was done by the service. Secondly, it's just plain less efficient. You need some 3-4 self-checkout machines and a dedicated person to watch them (to e.g. approve alcohol purchases), just to replace one clerk and their station, while keeping throughput more-less the same. What the stores do instead is, install 2 stations per replaced cashier, and have existing employees do the "watch duty" - which is why half the self-checkout machines end up being stuck for 5 minutes at a time, waiting for the overworked employee to finish resupplying a shelf, and walk all the way to the checkout arena to swipe their card a few times.

The queues get longer, people get more aggressive, everyone is doing unpaid work for the store. Madness.

[hasbot]

Heh, I find self checkout more efficient. I use multiple bags when bagging and pre-sort items based on where they go in my kitchen. Plus most baggers are terrible at bagging and just resort to a massive number of bags which kinda ignores the purpose of the bags.

[TeMPOraL]

I haven't been to shop with baggers, so that time is more-less constant for me (except for the pressure to bag faster, which isn't present with self-checkout). However, the cashiers are absurdly fast at scanning. Their workstation is optimized for this, and they scan through things faster than I can move them into bags. Even without any exceptional circumstances that make you wait for assistance at self-service checkout, scanning speed alone cuts the time per customer at least by half.

[zamnos]

On a per-item basis, absolutely, but what's the number of self-service checkouts at which they faster simply by having a much larger number of kiosks? Having 10 check stands open for 10 customers, each with their own employee is inefficient, but having 15 self service kiosks open isn't out of the question.

[bleep_bloop]

This. While not me personally, there is obviously a market (Japan comes to mind) for virtual, non-human based interactions and relationships.

While it seems pretty clear to me subjectively that removing human interaction from your life has a negative effect it's still desired by some and they're entitled to that.

[0zemp2c]

The very last thing shut-in men in Japan need is a SaaS virtual girlfriend

[EniasCailliau]

What if it helps them gain confidence? You could see a AI companion as a safe space to explore social interaction.

[bleep_bloop]

i don't disagree with you, but if you're already terminally online whats the difference.

The internet and social media has exacerbated a societal problem that has always existed, so now we have the problem why not make it more meaningful and less depressing.

[narrator]

The great thing about llama is it has no filter. You can ask it to act like it is a critical friend who will criticize everything you say to it. It can also be an overly compliant girlfriend. With some creativity, the degree that one can explore social interaction, one's own imagination, or one's own eccentricity on one's own machine in complete privacy without hurting anyone is amazing.

Being able to exercise ones mind like that completely outside the boundaries of normal social interaction with a emotionally tireless counterpart is kind of unusual and unique. It feels like we are at some strange point in history when all this stuff is blooming and before it gets shutdown. It makes me want to datahorde models!

The funny thing is that a lot of the comments here are saying, oh yeah, the big corps are going to make billions off of microtransactions pimping us virtual prostitutes that will also socially mold us the way the authorities demand. That makes me feel totally awesome as a unix geek who can figure out how to locally install a non-nerfed chatbot and talk to it about whatever I want, for however long I want, in complete privacy without paying anyone anything.

[echelon]

Amouranth is running one of these for fans at a rate of $1/minute [1, 2]. Supposedly already raking in money.

With that example, the flood gates are about to open.

[1] https://kotaku.com/twitch-streamer-kaitlyn-amouranth-siragus...

[2] https://www.polygon.com/23733515/amouranth-ai-chatbot-foreve...

[EniasCailliau]

Wow. Missed this update. Thanks for sharing!

[sgt101]

ahhh I remember the good old days when I could get 8% on some crypto investments and buy a nice ape for $4k before selling it for $400k if only I would wait for the right time.

[sebzim4500]

These people are selling a service, not soliciting investment.

Were you under the impression that the people paying to 'speak' to Amouranth are expecting a financial return?

[sgt101]

it's all grifting to me...

[qudat]

We have only scratched the surface with AI alignment. In particular, emotional alignment will likely have a big impact on our lives, maybe to the point of population collapse.

We are already seeing population decline and an “inverted pyramid” for developing countries.

https://bower.sh/in-love-with-a-ghost

[EGreg]

Have you stopped asking locals for directions because of Google Maps?

Did you go to the library or ask your parents stuff growing up, or did you google it?

Do you order in, get amazon delivery, watch netflix?

Our generation has already become like that.

https://news.ycombinator.com/item?id=35994494

[spigottoday]

lol, one day I was asked for direction from 3 different people walking in the wrong direction that were using googlemaps!

[Lapsa]

"dark showcase of what a lack of human interaction (..) can result in" - nowadays everyone and your dog wants to become a moderator. millions worth systems trying to automatically deduce your sentiment towards USA presidential elections, Russian-Ukrainian war and if you may be in a need of penis enlargement. I just quote Bender Rodriguez, Achmed the Dead Terrorist and filter out all the (unsurprisingly most of them) stupid places. vote with your legs and let them rot