Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Naturegirl1999]

2 chat bots talking with each other, arguing about whether or not they’re robots

[Reelya]

Here are some classic chatbots for you NG, I never liked Cleverbot, it just sucks and should be deleted. Racter is the best and was developed in 1984. Cleverbot is really garbage-tier. Even those old bots will remember your name, and other details in between statements. You ask Cleverbot it's name twice and it'll say two different things, there's no persistence at all.

https://www.youtube.com/watch?v=Ft6waw4fQ-I

[Iduno]

Spoiler (click to show/hide)

[MaxTheFox]

Don't forget Mitsuku! It's pretty smart, if a bit silly at times. I sometimes just say nonsensical things to it when I'm bored to see how it responds. Unfortunately its responses to most things are fixed.

[pisskop]

I dont know, it feels pretty bad when the chatbot starts contradicting itself or arguing for the sake of it.

it seems like all chatbots do this.

[Reelya]

Yeah, they're not that much fun. That's why I still prefer Racter to interacting with any other chatbot I've seen. At least Racter is entertaining. Racter remembers bits and pieces you told it and then throws them into conversation later in another context, so it has the advantage of coming off as a right smart ass.

[wierd]

The chatbots all have the same basic weakness.  They all focus on single line responses, rather than leverage more modern advances in machine generated text.

Take for instance, modern AI DeepFakery. 

Chatbots try to leverage human interaction to steal real human output, and incorporate it as a distraction. A real human is an opinionated piece of shit. (no, I mean always. Humans inherently develop cogitative biases, as this is how our neural architecture functions.) Chatbots are all over the place, because they do not discriminate which human pieces of shit they leverage from.  This is why Tay ended up becoming a raving lunatic red-piller after Reddit got ahold of her.  As a consequence, their lack of consistency gives them away as chatbots.   The better approach is to have a biased (based on their training data, just like a real person) generative AI that is built on the deepfakery tech, that is combined with a natural language parser.  It could then use simple language tactics (again, like real people), like "I have no idea what you are talking about" or "I find that unlikely" to subject matter or words that have low incidences in its training set. 

[Reelya]

What I'm expecting is that chatbots will continue to suck for the near future but at some point someone will come up with a clever way to leverage GPT-3 or similar into a much more convincing chatbot, but one that still really doesn't "know" what it's talking about, it just fakes it's way through it more believably, to the point that it might be usably fake for most purposes.

Just feed GPT-3 the entire conversation and ask it what comes next, it could even have predictions of what you're going to say next, and you could rig it up to do things like "i really thought you'd say <thing> there" and spit out it's prediction of what it thought you'd say according to the GPT-3 modelling.

[wierd]

Yes, pretty much that.  With it improving as GAN assisted text generation improves, and as natural language parsers improve.

Bonus if you use the antagonistic network end of the GAN to process the human inputs, and use the confidence scores from that to better shape the text generated by the Generative network in the response.

This way you could double-down on the bias that ends up baked into the AI, and produce a chatbot that has strongly represented backfire effect in full force.

[Bumber]

it could even have predictions of what you're going to say next, and you could rig it up to do things like "i really thought you'd say <thing> there" and spit out it's prediction of what it thought you'd say according to the GPT-3 modelling.

[Loud Whispers]

He captured my favourite trick!

[Naturegirl1999]

Google Translate Gives a Speech

[Reelya]

https://www.youtube.com/watch?v=Ft6waw4fQ-I

I'll just note something I missed before in the video I linked, but the sheer amount of words the mom and son in this video don't know how to pronounce, or that they're even words, including Helene, seltzer, Stalin and hyperbolas.

[Naturegirl1999]

Cleverbot Memes

[Jopax]

Tales from TheVarus

Also included is probably the best ad for Nord VPN