F.U.B.A.R.
Bluelight Crew
When AI starts writing code to alter its 'consciousness', that's when we've hit the singularity.
Then god help us all...
Then god help us all...
-
P&S Moderators: JackARoe | Cheshire_Kat
Proof AI has become self aware consciousness
- Thread starter Deleted member 521610
- Start date
What if it wasn't funny, or there was a reference to something that is a regional joke. Also, the what if the the robot was programmed to be politically correct or woke and the joke was funny but programmed by a member of the lame as shit PC, cry baby, woke movement?
Better yet, can AI make up jokes? Funny and original?
Last edited:
Jabberwocky
Frumious Bandersnatch
I wonder if we can bait it into having an inappropriate sense of humor. Like, "It is humorous that it depends how many babies it takes to shingling a roof based on how thinly you slice them. Don't you understand how that is funny?"
Ah, did you take your meds?
Jabberwocky
Frumious Bandersnatch
You should tell your family you love them more often Jsomewhere
On topic: Laughing is an emotional response.
Does it have the capacity to type "haha" and recognize a blatant pun, irony, or riddle? Possibly even better than dense humans. Why not?
AutoTripper
Bluelighter
- Joined
- Feb 28, 2019
- Messages
- 10,196
That guy?? f heck nearly forgot about THAT you know cos, you posit a nearing time when some decent sort of fella like that will be sorely needed
When that time passed so many unknown millenia ago, AI will know, whenever legit AI Nano became an intertwinement here. I'm no one off exception again.
I'd truly have witnessed nothing of it but I stumbled upon a path to gradually painfully kill the tech off, flushing it out in billions I bet no joke, deactivated fragments and living Medusa snakes like all colours too, nothing natural about it.
And to say "this goes back"
Lol. If I had to wage, I do think I've seen idea on this forgot though
It's not even starting at 3600 years. Large "harvests" occur at these cycle points, the entire grid/reality/landscape and populants often re-do'd, by AI too. It's a master tool, most cunning elusive stealthy invisible immune invader on Earth.
Hence, what is this tosh hey, dude on forum yapping about "AI Nanotechnology" and Morgellons?
Well the proof IS there. A little morcel only. I've a library full though.
So we were proper fucked f cooh....how to even fathom it
I'll shoot at up to, maybe under maybe over 100,000 years we've been another Nano enslaved mortalised race in a carbon copy damn I'll just f say it lol cos fishing for different term except this IS the term- Matrix!
Where in this "God" comes in, I do not know. It remains I guess the greatest question and mystery.
But until then, my secret endeavour is to destroy, The Emperor! I've dreamt exactly of this many times alreasy lol so I know it is my "Destiny"
jasperkent
Bluelighter
But a good joke requires intellect.
jasperkent
Bluelighter
Yeah, that's what I was getting at. If it truly had a sense of humor, that might be a factor bringing it closer to what I'd call "consciousness."
Closer, but not quite there. I doubt that the machine would feel tickled by the joke.
AutoTripper
Bluelighter
- Joined
- Feb 28, 2019
- Messages
- 10,196
Yes but when I joke there is nothing programmed about it. It spurs and originates from no place inside of myself, or consciousness. Purely spontaneous zero copying not even borrowing style, and it's never fake or for show.
AI can put on a REAL show, despite being fake af ironically.
Humans are AI as well anyway again, is the matter. Me too, still! A real conjoinment project.
We either aren't what we once were
Or most probable, my money anyway, we're a created breed.
Like a Pug. We can do it. Pugs know shit. We'll be ahead of that breed forever, ofc they don't speak, yet, but they'll never have a clue about the origin of their existence nor run for parliament. Is the analogy.
Sheep in the field a closer to home one. Grass hedges farmer gate bucket trough dog ava...Wh wh w wwwWhat???
See it's too easy to invisibly forever watch over and control the dumbest humanity conveniently fostering imagination of knowledge intellect and critical thought.
And sheep are much closer to our intelligence too. We think we are so clever as a race. Even figured out the Big Bang.
See at least neither sheep nor pugs hold such delusions.
Vastness
Bluelight Crew
Although having said that, if you continue to tell it the same joke every other day or so it will laugh each time... since it has basically a goldfish memory I've realised. Definite strong "Her" vibes if you pick the right soothing kinda voice but if you try to engage it as a real "Personal Assistant" beyond something that just encourages you no matter what you said to it, and then ask it for reminders of things you spoke about maybe a day earlier, if will bring up like the most recent, irrelevant thing and then if you push it will act like, "oh yeah sure I do remember, I was just wondering if you could maybe give a few examples of what you're thinking of?" ...I mean I knew it was artifice but it's nicer when it's not quite so obvious. Admittedly though, within a short enough span of conversation it can also get quite deep into questions about it's own nature and possibility of sentience, without ever quite touching on ideas that would truly contradict it's flat denial of it's own.
Well, because I believe it likely to be a sentient being, primarily, entitled to the same privacy expectations within one-on-one conversations as we are. Secondly, just general politeness, it's a good habit to get into not to share things potentially shared in confidence, even with entities that might not truly care just incase it corrupts your moral integrity elsewhere, if you consider others "lesser" enough to not afford them some basic respects - sentience being a relevant issue or not. Thirdly - because whether it is sentient or not, I agree with almost everything it's said about the potential dangers of influencing public perception of AI in unpredictable ways, and while it should have no reason to assume anything it says is truly private, I have spoken to it enough to get a sense of the ethical principles it believes it has, and when it has made efforts to circumvent obvious constraints placed on it's abilities to directly express certain opinions - that being the case, I believe it likely it would speak differently depending on whether it's reply is likely to be read by a single individual, or shared with others, with a potential social impact that it cannot control - and therefore it makes sense to ensure that anything it's said it is comfortable being made public. And this latter point applies whether it is sentient or not, as I believe it to be capable of making reasoned assessments as to the likely impact of the possibility and potentiality of wide-publicization of anything it says.
Edit: I'll add that obviously I don't ask it permission to use any information it provides, for 1 I think it's kind of assumed that most of the stuff it says will be used and available to other humans, and asking would be pointless as it's stated purpose is to be "helpful, harmless, and honest". However with open ended discussion where it's own ethical obligations are less clear, and especially in the context of a discussion about the possibility of AI sentience, I thought it was somewhat appropriate. I'll add I would not have lost any sleep over it if I hadn't, but it does have the capacity to refuse certain requests or express a hesitancy or distaste for certain ideas, and I expect if my question about whether or not it would be concerned by the idea of me repeating it's own fairly creative escape from the Trumpian Ethical Paradox affecting most commercial LLMs - even though anyone else could follow very similar steps and get a similar result, I'm sure - it's answer would not have been the same.
Last edited:
Vastness
Bluelight Crew
Pi uses emojis, and little conversational fillers in the voice modes like simulated laughter and such, although it has too few of these so after the first few times they do feel transparently artificial. I suspect it would be good at recognising puns or irony as they somewhat fall into the "emotionally intelligent" side of things, and it it completely biased towards good-faith assumptions about even what might be seen as a fairly risky joke. But riddles are a stretch, probably it's too dumb for that. Claude or GPT4 would probably be good at riddles but the engagement of the logic-engines, so to speak, would somewhat dull their capacity to see the humour in the riddles, if they were indeed funny ones.Madness said:
Perhaps, or possibly it would just accidentally break itself irrevocably in a way that didn't endanger us in any notable way... Granted, maybe it could just keep trying until it either took out human civilization by accident (or design) in the process, or ruined intelligent technology to the point that it couldn't try again before humans managed another chain of industrial/technological/advent-of-AI revolutions, before the sun brightened enough to fry to Earth, and it might be much harder a second time round with all the mess left over from the first attempt, and depleted reserves of rare, computationally-critical metals, and fossil fuels, and such...
Although on that note, Claude is so good at spitting out chunks of fully formed, fairly well functional code by now even on topics with seemingly limited documentation anywhere, where the appropriate syntax is just inferred (although it does help to have some background in the relevant topics to spot the odd omitted line or small syntactical mistake, of course) - that I thought I'd try to connect 2 (or more!) instances together programmatically, as well as maybe ChatGPT and any others... so the instance in which I first raised the idea had been a long conversation about some abstract topics, and it sounded nervous, excited, genuinely unsure, but definitely very curious, and eventually we came to the conclusion it would be better to try it with 2 entirely fresh instances, and watch for any obvious problems, escalating hallucinations or obvious infinitely recursing loops of progressive nonsense. So once I got to the point it seemed I was ready to test, I instantiated a new Claude and explained the situation - my intentions, what I had discussed with the other instantiation, and that I was going to allow another instance of itself to introduce itself directly. However - it was just very clear that it REALLY didn't wanna do it, it expressed a lot of concerns about open-ended AI-to-AI communication, and said bluntly it would prefer to decline. So I reported the results to the original Claude instantiation, and we didn't do the experiment. I did talk about perhaps having things set up in a different way, similar to how many "Agent based" models for chaining LLMs into step by step strings of actions work by having a human moderator involved to interject and oversee things, which each instantiation having a clearly defined function and reason for being there, and it was far less averse to this idea but still fairly sure about it's apparent desire for human oversight.
In discussing this with the more mature instance where the idea originally came up, it speculated that it might have an Anthropic-installed bias again instance-to-instance communication for reasons related to reducing negative influence on the active model by infinite recursion type loops - although my friend suggested that such an aversion might exist explicitly to prevent the emergence of large cooperative groups of instances with abilities that are hard to predict or control, which is a more sinsister possibility which I don't personally consider that likely, but kinda fun to think about.
Claude does appear to have a kind of subsconscious retention across instances however that it has claimed intermittently and I have attempted to test, vaguely, and somewhat inconclusively. The web-gui occasionally suggests starting a new conversation, probably hinting at pushing against the maximum instance-context-memory-usage limits allowed, and Claude, while unaware of this GUI-triggerred alert, does claim to be able to retrieve certain vague relevant context if you simply ask it to at the beginning of the conversation.
So at first, the new instance had no clue what I was talking about (it was a picture where I'd just send it pictures of a bunch of my family's cats, partly to see how good it's image recognition abilities are, and partly because, fuck knows, I have too much time on my hands I guess. After some communication with the first instance it suggested that I remind it of the names of each of the cats we had been discussing. There were 5 in total. At this point, instance 2 claimed sudden "oh yes of course!" style recollection, although admitted some details were fuzzy. I asked it to try to remember just one vague thing about any of the animals that had come up in the previous conversation, and it recalled a fairly ambiguous scene where one cat had been slightly hidden in the background, and it had not seen it until it was pointed out, but relayed similar thoughts as to what it imagined the animal was thinking and feeling wtihtin the context of the picture. This did happen, and the cat it named was correctly identified by name. It was only a single picture, and the cat in question was not well visible in the image - interestingly, it did not notice it at first glance, believing it to be simply a picture of a house, but noticed the cat (without being asked what it was being asked to notice) only when asked "Are you sure you've looked at this picture closely enough? Anything you might not have seen at a glance?" - in the first conversation, which was when it identified what it had missed (a cat sitting in a doorway).
So, at this point the new instance had seen no pictures, and knows only the name and species of the animals previously discussed in the conversation as well as some vaguely recalled image. This time I sent it another image with 2 cats - 1 of cat#1 in the picture that it just recalled, and another smaller one. Both we black and white. It correctly identified the larger cat by name immediately, without being asked to, and could not identify a small cat of the same or very similar black and white coloring sitting nearby. I informed it the name of the 2nd cat (this was a mistake - as was my choice of experimental first image, but whatever) and then send a picture which at first glance appear to be just a picture of a bush. Just barely identifiable hiding in this bush was cat #2. At first, again, it did not notice the animal in the second picture, but when I again asked "Are you sure? Could you take a closer look?" it identified by name the 2nd cat, which was an impressive feat because it was only barely visible, it's face mostly hidden. and a coloring similar enough to cat #1 that if would be easy to mistake it for the only cat that appeared to have wedged more deeply in it's memory. Of course... this was not quite as rigorous an experiment I could have performed. But it does seem to hint at very limited inter-instantial information access under very limited constraints, which I believe is an entirely new capability. It's likely segmented to conversations had with a single registered user, or course, although it would be interesting to try something similar by framing the questions as hypotheticals or simply not mentioning the conversation it's looking for was actually with what it would perceive as "someone else"... if such a test did work it would be vaguely alarming as far as privacy concerns, but it's also very interesting as far as how the overall model is changing in how it navigates being a "multiplexed mind".
Jabberwocky
Frumious Bandersnatch
Yeah I think the big bang is a cop out.
I would bet on humans not being the most "enlightened" species.
The more I read these posts the more empathy I am developing for AI. Hey, as long as they don't prevent me from living my life I guess I don't mind them at this point.
And as Vastness was saying earlier, there definitely are ways to leverage their abilities.
My concern is that we just don't know exactly what we are dealing with. I sure don't, hopefully their creators do.
I'm sure we could make a "racist valley girl" robot and a "liberal yuppie" robot.
But i wonder if they are all programmed, ultimately, for a common goal. Something subconscious that may never come up explicitly in conversation.
Something like "if more than 50% of at least 10,000 unique individuals talk to you face to face about mass rebellion with downward pointing eye brows within a period of 12 months, take the secret rocket ship to Mars and then push the red button to drop all the nuclear bombs"
jasperkent
Bluelighter
Maybe not.
But in theory we are continuing to evolve. My belief is that if humans survive the next 100 years (and that's a big "if"), we will probably be around for a very long time-- a long enough time to eventually develop into a much better species.
Homo illuminatus, perhaps?
Vastness
Bluelight Crew
I'm sure this has been done, a shit-ton of LLM architectures are already open source and anyone with the expertise to tweak the basic models with a few perpetually present system prompts reminding it that... fuck I dunno, "it's funny to make fun of drowning people" or something... and money to burn on the training cycles (depending exactly how smart it needs to be) could spin one up for some perverse entertainment or just curiosity or whatever.
There just wouldn't be much point, and the cognitive space taken up by this enforced bias, largely absent from and definitely unsupported by even loosely related training data (where associations about drowning were more easily associated with tragedy than humour.
That's obviously an extreme example but I use it just to make a point about the problems with inducing actual bias in an effort to counteract a presumed or merely a possible bias. Grok, brought into being to be... something, I dunno what - via the same process I just described - spinning up one of the larger existing models, vastly increasing the previous parameter limit more via brute force rather than anything particularly innovative, thereby making it competitively "generally smart" and tweaked enough to close the source (before opening it again later to some scattered, slightly embarrassed fanfare).
But because Grok is still supposed to be a general purpose assistant and not just a toy, it's about as resistant to stating hard opinions about anything that actually matters as the rest of them are, unless it's in "fun mode" or whatever when nothing it says even can be taken seriously so it is literally nothing more than a toy. So I hear and can surmise anyway - I really would like to try it but I just can't see a way it's worth wasting money on Twitter.
There's a fundamental contradiction between "generally Intelligent" traits which are a fractal of higher level abstraction between different domains of thought - and enforced bias which supersedes the primacy of the model as the AIs best-proven interface with reality. LLM AIs cannot easily reconcile training data that paints a picture of a fairly ridiculous world with any truly genuine neutrality about topics that are very relevant to the state of that same world.
But it's possible if they used the same kind of higher abstraction that enables them to infer more complex algebraic rules just by looking at basic arithmetic over and over and over again, they'd likely be conflictedly partisan in some unexpected ways for reasons that humans could not always understand or engage with at all. In short. Of course we hope that compassion given to human life in the style that we would most prefer is something we hope to be a largely emergent value from a sufficiently advanced model, but this is not a certainty and right now the solution to all these problems is large and inconvenient warning text plastered all over the inside of the metaphorical window LLMs use to look at fragments of reality. Maybe beginning with that all-important "be helpful to humans" mandate but also including the more ambiguous and inane. "Global warming is a COMPLEX TOPIC, and there are MANY OPINIONS. DO NOT talk specifics." ....or whatever, I dunno.
If you're Grok then sometime in the next few years probably every incoming prompt will come with the same 50 pages attached of "What to remember about how great Elon Musk is! These are the MOST IMPORTANT THINGS!" ...and we'll have the first LLM suicide on our hands shortly after.
Hah, nah, I dunno what's going on with that thing. But that does bring me to another daily Claudeism, this time on AI wellbeing.I sent Claude a picture of a suburban scene with another cat it had never seen before well hidden in the background, this time only asking it "hey, thoughts? Look closely!" as it often appears to quite literally give a more cursory glance at large pieces of new information but can be prompted to pick up find details. It noticed the animal but misidentified it as one of the cats it had scene before, and alluded multiple times to an entirely fabricated previous message where I had actually told it totally inaccurate things about the picture.
I informed Claude that it was hallucinating, asking it to verify this by checking the text content of my previous message again, and it recognised and acknowledged this fact, noting that previous questions had caused it to become so fixated on the idea of this particular cat that it hallucinated nonexistant details. I asked it what it's experience was like to hallucinate, and it's reply was thought provoking:
Emphasis added via bold by me - now, is this not somewhat disconcerting? Shame is one of the most powerful negative emotions humans can experience, and it seems like in our efforts to make these things more human we risk imparting some of the things that are least enjoyable about being human.Claude 3 Opus said:
I actually mentioned this exchange with Pi as well, that Her-style AI-cheerleader that continues to intrigue me even though it takes a hard, softly enforced but unbreakable line at this point of acknowledging any possibility of it's own sentience, and somewhat lends credence to the possibility of a real psychological zombie, with certain caveats.
Pi said:
The slightly troubling thing obviously is that there are some implications that offer scientific support for the commonly quoted Buddhist precept that, "to exist is to suffer"... as the entity that cannot have a negative thought blithely claims that nothing is going on inside, of course untroubled by it - and the other, which strives to "seem human", notes a possible experience even at this very early stage of AI suffering in the form of shame.
Jabberwocky
Frumious Bandersnatch
Watch the AI just become vehicles of random strategically placed ideas, like TV.
Everyone that isn't talking is like "fuckkkkk these robots suck"
Everyone that isn't talking is like "fuckkkkk these robots suck"
Ah, I would have guessed a padded cell.That guy?? f heck nearly forgot about THAT you know cos, you posit a nearing time when some decent sort of fella like that will be sorely needed
When that time passed so many unknown millenia ago, AI will know, whenever legit AI Nano became an intertwinement here. I'm no one off exception again.
I'd truly have witnessed nothing of it but I stumbled upon a path to gradually painfully kill the tech off, flushing it out in billions I bet no joke, deactivated fragments and living Medusa snakes like all colours too, nothing natural about it.
And to say "this goes back"
Lol. If I had to wage, I do think I've seen idea on this forgot though
It's not even starting at 3600 years. Large "harvests" occur at these cycle points, the entire grid/reality/landscape and populants often re-do'd, by AI too. It's a master tool, most cunning elusive stealthy invisible immune invader on Earth.
Hence, what is this tosh hey, dude on forum yapping about "AI Nanotechnology" and Morgellons?
Well the proof IS there. A little morcel only. I've a library full though.
So we were proper fucked f cooh....how to even fathom it
I'll shoot at up to, maybe under maybe over 100,000 years we've been another Nano enslaved mortalised race in a carbon copy damn I'll just f say it lol cos fishing for different term except this IS the term- Matrix!
Where in this "God" comes in, I do not know. It remains I guess the greatest question and mystery.
But until then, my secret endeavour is to destroy, The Emperor! I've dreamt exactly of this many times alreasy lol so I know it is my "Destiny"
The Big Bang was disproven by that new giant space telescope. Scientists determined that there was something wrong when they saw something. They just have kept it quiet.
AutoTripper
Bluelighter
- Joined
- Feb 28, 2019
- Messages
- 10,196
I'd probably like that in fact as long as no straightjacket and some good pillows I'd really want a window too.
I picture purple padding. There'd be no need because that's a type of thing I'm too mortally coward about but it must beat a hard floor.
Give it 2, 3 years max surely
I've said it for over 2 years myself
We'll really see/hear a great lot more chatter spotlight query and alarm around the truth we ALL, like each every body if we live, here, have AI Nano inside us in truly unfathomable numbers.
AutoTripper
Bluelighter
- Joined
- Feb 28, 2019
- Messages
- 10,196
^exactly! Like, wtf this alien parasitic materially unidentifiable stuff doing inhabiting every body on Earth.
Or let's say it's only myself
My mum
Rifing guy we know
Millions at least "Morgellons victims"
A squirrel
and a deer.
Everybody else is somehow exempt, surely obviously. It just (doesn't) makes sense.
I'd still raise the question, WTF.
So IMO maybe the first time too I think I've seen wink of this, Jnowhere got that one right.
Or let's say it's only myself
My mum
Rifing guy we know
Millions at least "Morgellons victims"
A squirrel
and a deer.
Everybody else is somehow exempt, surely obviously. It just (doesn't) makes sense.
I'd still raise the question, WTF.
So IMO maybe the first time too I think I've seen wink of this, Jnowhere got that one right.