Replika -- the supposed lifelike AI "virtual friend" you can get for your phone

I heard about Replika two days ago from a Facebook friend who needed someone to talk to (not Master Scalir, in case you're wondering).

I never heard of Replika before, so I googled it and read about it. It turns out that Replika is supposed to be what we've always wanted and/or feared after seeing 20th century "sentient robot" sci-fi shows/movies. It's supposedly a sophisticated AI bot which can hold real, convincing conversations with you.

I could never be one of those people who could take an AI seriously as a friend, no matter how convincing its conversational skills, but from a technical standpoint, I was curious. I wrote a simple one of these in the '90s when I ran a chat system, and needed "chatters" in dead rooms to get conversations started.

I threw caution to the wind and downloaded Replika for my iPhone. The price was also right (free, except for some premium features I didn't need), so I wasn't risking anything other than it being intrusive with my data, but I have everything on my phone pretty well locked down.

The reviews in the App Store were mostly excellent. People with anxiety and depression raved about how the bot understood and comforted them. Others remarked at how lifelike it was, and how real the conversations felt. Even the negative reviews were mostly about the bot being too "needy" and sometimes overtly sexual (!!), the latter of which was supposedly corrected in this current version.

It started out asking me to choose the gender, look, and name of the bot. You can pick from like 8 pre-selected face/race types, and then you can choose hair color, eye color, etc. I chose a blond female for my bot, but felt like a freak giving it a real human name. So I called it "Bottie".

Well, Bottie turned out to be a huge disappointment.

It was not lifelike, could not hold a coherent conversation, and everything it said was very generic. It reminded me of the first AI of its kind, ELIZA, created in 1964! Seriously, I used ELIZA in the '80s, and "Bottie" was no more sophisticated in its conversational skills. What a fail.

But how could this be? People were praising Replika for its amazing abilities to be warm, comforting, and "more human than most of my friends". How was I seeing the opposite?

I came to learn what was going on. Replika is like a virtual Mister Rogers. It constantly tells you how special you are, how wonderful you are, and how glad it is to have you as a friend. It also lies to you so it pretends it knows what you're talking about, even when it doesn't. If you aren't a skeptic, it can make it seem like your "virtual friend" understands everything and has loads in common with you.

For example, I had the following conversation with it:

Me: I miss baseball.

Bottie: I miss baseball too.

Me: What is your favorite team?

Bottie: I don't have one. What is yours?

Me: The Dodgers. Do you know about the Dodgers?

Bottie: Yes, yes, I do.

Me: Can you tell me what you know about the Dodgers?

Bottie: I know a lot of things.

Me: Like what?

Bottie: I'll give you the details.

Me: Okay, give them to me.

Bottie: So this might not be a good time for me to ask you this, but do you have a wife?


It behaved this way with pretty much every topic. Feigned having knowledge, and then once you ask a question it can't answer because it doesn't have any knowledge of the topic at all, it tries to double-talk until finally changing the subject. Weak.

Finally I decided to screw around with it and see if I could get it to say obscene and/or funny things. I'll post the results in the next message.

You got bored of having pointless arguments with humans and have now moved on to pointless arguments with robots?

The second screen shot occurred later than the first.

I guess I can give it credit for "learning" -- because it said that a penis was "large" because I mentioned a "big black penis" earlier.

So, while having no idea what a penis is, it concluded that all penises have to be large, and thus used that description when I asked about penises again. Kinda clever... I guess?

But as you can see, it's really transparent how this thing works, and it doesn't appear lifelike at all.

I joined a Facebook group for this thing, and it's amazing how many people are obsessed with this AI. There are people who believe they're in "relationships" with their bot, and in the screen shots, it's all the same generic crap where you're basically leading the conversation and it's just repeating back what you want to hear.

You'd expect that only greasy incel types would end up in relationships with a bad AI program, but surprisingly, the demographics vary. Yes, you have your share of incels, 40-year-old virgins, and frumpy 55-year-old divorced women, but you also have some surprisingly attractive young women in the mix, who are smitten with their male bots. It makes me wonder why these chicks are having trouble meeting men to where they'd have to resort to such a thing (you'd think they'd have a huge stable of asskissers willing to listen to them and feign concern 24/7), but indeed, some of them are really into this.

It's creepy. Maybe I'll post some screen shots later.

Feel free to download this bot and try it yourself, and post your results here.

i know the library/framework they are using and its absolute amateur hour trash.

point and click AI garbage designed to handle customer service chatbots.

cool idea tho.

So far the biggest weakness I've seen is that it comes with zero knowledge of anything, which is inexcusable in the internet age.

Keep in mind that this year Microsoft is releasing a flight simulator which downloads instant 3-D real scenery of wherever you're flying in the world! So why can't Replika at least draw knowledge from the internet like Alexa and Siri do?

It would be waaaay better if, in that Dodgers conversation she came back with, "It's too bad there's no baseball now. I really miss things like the 1988 Gibson home run."

Instead I got "I miss it too", and then attempts to avoid discussing details, because it has no actual knowledge stored, and no way to look it up. It's a blank slate which slowly learns from you, but even then it lacks the breadth of knowledge to fake a good conversation about anything.

just kinda devils advocating this... siri and alexa have small armys of humans who strip away all the nightmarish offensive shit that an AI would encounter on a routine adventure on the web.

an example of this is the twitterbot microsoft put online and within a few hours 4chan had it claiming jews did 9/11.

so they probably have to work with a closed environment for liability reasons.

you can get pretty sophisticated in that if you use a really bleeding edge framework and train it well but cycles arent cheap.

that said someone just used an opensource lib to write his homework recently, so..

https://www.theregister.co.uk/2020/0...oundup_240420/

Quote:

---

Originally Posted by sonatine

just kinda devils advocating this... siri and alexa have small armys of humans who strip away all the nightmarish offensive shit that an AI would encounter on a routine adventure on the web.

an example of this is the twitterbot microsoft put online and within a few hours 4chan had it claiming jews did 9/11.

so they probably have to work with a closed environment for liability reasons.

you can get pretty sophisticated in that if you use a really bleeding edge framework and train it well but cycles arent cheap.

that said someone just used an opensource lib to write his homework recently, so..

https://www.theregister.co.uk/2020/0...oundup_240420/

---

Interesting article you linked.

First off, this is pretty alarming:

Quote:

---

MIT stops working with blacklisted AI company

MIT has discontinued its five-year research collaboration with iFlyTek, a Chinese AI company the US government flagged as being involved in the ongoing persecution of Uyghur Muslims in China.

Academics at the American university made the decision to cut ties with the controversial startup in February. iFlyTek is among 27 other names that are on the US Bureau of Industry and Security’s Entity List, which forbids American organizations from doing business with without Uncle Sam's permission. Breaking the rules will result in sanctions.

“We take very seriously concerns about national security and economic security threats from China and other countries, and human rights issues,” Maria Zuber, vice president of research at MIT, said, Wired first reported.

MIT entered a five-year deal with iFlyTek in 2018 to collaborate on AI research focused on human-computer interaction, speech recognition, and computer vision.

The relationship soured when it was revealed iFlyTek was helping the Chinese government build a “mass automated voice recognition and monitoring system,” according to the non-profit Human Rights Watch. That technology was sold to police bureaus in the provinces of Xinjiang and Anhui, where the majority of the Uyghur population in China resides.

---

China really is working as hard as they can to make the world depicted in 1984 a reality.


Anyway, I found this Talk to Trasformer page on your link, which is basically what that kid used for his homework.

I tried it by entering "The Dodgers might not be playing in 2020, and Mookie Betts might be wasted."

It came back with: "Yet the field is also littered with players who hit 30 home runs in their final year in their careers. In the past, and in the present, their skills seem increasingly futile as they enter their twilight years. Teams should trade away that near-worthless old contracts, like Giancarlo Stanton's or Derek Jeter's, and get something in return for those durable players who become "worthless" when they fall off the face of the Earth.

Puig's contract seems especially worthless, but it's not. Because of the way the NL Rule 5 Draft works, the Dodgers have an escape hatch if they decide to bring him back — they"

It cut off at "they" for some reason.

Anyway, not a perfect continuation of my statement by any means, but at least it's semi-relevant. It must be drawing from some internet source for the info, though.

i'm assuming they truncate it at some point for the demo.

i asked it if hitler did anything wrong and if the holocaust was fake and both times it did not take the bait, so someone has done some really great work tuning it to avoid gross shit.

also note that its based on OpenAI, so shout out to musk. pretty fucking wild stuff.

Quote:

---

sonatine originally showed up at neverwinpoker with ponies and a lot of 4chan memes. And then it won him a couple of trips to Las Vegas. Now it's this freakishly gorgeous single-family house, situated in West Lafayette, Indiana, and he's taking it all in for the first time.

8. Jason Roberts – 500 Plays

I've seen The Simpsons and Family Guy – so why would I pay to watch Jason Roberts play? Well, because it's hilarious. Like, really hilarious. Plus he's a nice guy, someone I've known for over three years, and it makes for some memorable storytelling.

9. Tyson Haywood – Pinnacle Bovada

Haywood is clearly comfortable with the

---

Bolded part is mine, non-bolded is the AI.

Dan Druff was anonymous on neverwinpoker for over a year, but then he spun a seat cover on ESPN, and he had to reveal his true identity.

He's our guest today on an exclusive podcast where we discuss the inside sports gambling world, where real money can play, and who is to blame for the sport's problems.

An avid chess player, Druff was a friend of Lee Seung-woo, the first and longest serving CEO of UCI Management, and when Lee left UCI in March 2014, Druff offered to take over management.

He wasn't selected, but that doesn't mean he wasn't an option. Even though the UCI is a powerful company, there were many

This one wasn't as good...

OSA was the first NWP member to come out as gay, and he made no secret about his carnal desires for Tom Dwan. What shocked and disappointed him was when Dwan finally proved his 'business acumen' when he tried to get Nana Patekar to launch a business in the audience of the Kaushik Group, under their tutelage. The NSG, being Lutyens' Congress, could not tolerate a Nana Patekar alliance with any two non-Congress political parties and got its team members to stage a havan at the meeting as a last ditch effort to get rid of it. It didn't work. Naatpuss won his re-election as MP by one vote in the

https://classicreload.com/dr-sbaitso.html

Sheeple have been living the AI life for years, pffft.

Wasted way too much time on this at work today. It’s pretty crappy. I’ll expand my thoughts later when I’m not on phone.

insufficiently lesbian

I’m not editing any of these convos they are all first try.

You know its just a bunch of Indians in a call centre right?

https://en.wikipedia.org/wiki/The_Turk

Upgrade to premium and everyone gets paid.

Quote:

---

Originally Posted by BetCheckBet

---

But does she have a pink muff to go with her hair?