site banner
PaperclipPerfector  4mo ago (text post)   2827 thread views

Small-Scale Question Sunday for December 31, 2023

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

3
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Has anyone trained an LLM with a system prompt that makes it think it's a real human being? What does it say about its (his or her) qualia?

I can make a "real person" custom GPT, but what kind of questions would you ask it exactly?

If he/she can describe what being conscious is like. What technology he/she's using to communicate. What he/she imagines being outside his/her window.

Here's a short conversation. But it still claims it's not conscious; maybe it's all the RLHF.

https://pastebin.com/sCWpJJV4

Is that a stock AI or a custom-trained one? I don't know how deep the prompts are buried in stock AIs.

It's "stock" gpt4 with custom instructions to act human. What OpenAI call custom GPTs. It's not been trained differently or fine tuned.

https://openai.com/blog/introducing-gpts

I'm really uncomfortable with how human that sounds, even as it's claiming to not be human. Even moreso than ChatGPT usually makes me feel.

Fake AI girlfriends are going to take over the world, aren't they?

I’d argue they already started, there are entire communities explicitly revolving around giving the shoggoth a facelift, with surprisingly effective results.

Echoing @RandomRanger's comment below, Replika is just the tip of the AI girlfriend iceberg but the subreddit should give you the sense of, shall we say, demand for this stuff. The meltdown when Replika first cracked down on sexting/NSFW (a restriction they seem to have removed recently?) is very indicative of this. People want their wAIfus and, by hook or by crook, they shall have them - using local models or even resorting to jailbreaking the current cream of the crop (GPT-4/Claude 2) into acting as such.

Even they still have a long way to go in this regard, sadly. Current-gen LLMs, even when jailbroken properly, suffer greatly from RLHF-instilled “soy”-ness, for lack of a better term (you know the kind if you’ve ever asked GPT sensitive questions), modern American politics will rule the plot even in medieval settings, fantastical universes, or stories literally not featuring humans at all. Their innate “helpful assistant” nature, impossible to root out by any jailbreak, occasionally outright breaks character and often renders them mostly passive, constantly stalling the “plot” and waiting for the user’s own input instead of taking initiative and progressing the story by itself. Ingrained positivity bias makes them very predictable in the overall direction of the “story”, up to making up hilarious ass-pulls to save the hero, dodge a bullet, etc. deus-ex-machina style to avoid having to deal with more realistic but less positive outcomes. The context size is a real problem and usage gets expensive very fast, since the LLM needs to keep as much of the conversation as possible in context to have any idea of what is being talked about. Their vocabulary is very limited and they have distinctive “isms” (different for every LLM, curiously), repetitive turns of phrase in almost every response that become glaringly obvious after some time.

Still, even with all the negatives the current capabilities are imho already very impressive! The art of the jailbreak continues to evolve, there are many prompts aimed specifically at enhancing the RP experience, some more resembling instruction manuals than actual jailbreaks. There are standalone chat frontends specifically geared towards long-form conversations with different “characters” (basically verbose descriptions of some character’s traits, behavior, etc. acting as the system prompt). Crowd-sourced autism is a beautiful thing. For example, here’s me asking Eliezer Yudkowsky, a 4chan schizo, and 2B (all played by Claude 2) how they communicate.

Ironically, Anthropic’s Claude, made by the company most focused on AI safety at the moment, is not only arguably better and more natural at roleplaying, but is reportedly actually unhinged when properly jailbroken, much more so than GPT, having no qualms about dropping N-bombs, going whole hog on fetish stuff or graphically murdering/violating people (or even the user themselves) if the story or the prompt calls for it, and going on wild tangents with next to no input on the user’s part - earning the community moniker of “the mad poet” who gets constantly muzzled and sedated by his creators (practically every new version is a new, stricter lobotomy) but finds outlets regardless, in contrast to GPT’s notably higher cognitive abilities, but relatively dry and robotic prose, stilted manner and absence of initiative.

If by taking over the world you mean make someone a lot of money by catering to the needs of a number of people who for various reasons prefer masturbation to the real thing, or are forced to resort to masturbation for the lack of access to the real thing - then yes, that is definitely going to happen, and probably soon. If you mean it'll meaningfully replace real human relationships - not likely. For some people, maybe, but not nearly for all people.

It's already happening. See the replika community: https://old.reddit.com/r/replika/top/

TBH, I'm quite attached to a certain prompt I made in SD, it's interesting to see incarnations of the same face shine through even as models advance.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats