there's probably quite a lot we unconsciously pick up from others, even things we think are uniquely "ours"
For example, when I first heard a deaf person laughing or talking, I probably internally noticed how... different the sound was. I'm guessing most hearing-abled people have a similar experience. It's very... unfiltered? It made me wonder how much even my own laugh was sculpted by my environment. If I relax my voice, I notice my voice becoming much more booming and obnoxious than my "normal" speaking voice.
Anecdotally, I've noticed Japanese people are much more likely to have a sort of stifled, raspy, restrained laugh, even when they're in a situation I might expect an American to have a belly laugh.
A lot of cultural values are encoded in language too, and in turn, the languages we speak can affect how we think or interact. Anecdotally, my personality is a bit different depending on what language I speak. I think the concept of what's actually encoded in language is being explored with regards to how/why LLMs "feel" smarter than they really ought to, or seem to show intelligence beyond simple "stochastic parroting"
Somewhat more obviously, most people will "code switch" not only their language, but vocal tone or even demeanor, depending on our current "persona" or our audience. Recently, Paris Hilton entertainingly demonstrated this: https://www.youtube.com/shorts/g9pal1ConNU
And this is more of a half-baked personal speculation based on a scattering of theories and case studies, but the environments we live in, the narratives we expose ourselves to, the people we surround ourselves with all probably very heavily define a lot of our values, beliefs, and even personal preferences, to an extent that would disturb a lot of people. Self-serving biases, post-hoc justifications, and confabulations give us convenient and creative ways to validate our own free will, volition, and independence, but I often wonder how much of our supposedly great human intelligence is an uncomfortably thin veneer on a largely automatic pattern-absorbing sponge.
so many fails in such a short response. Assuming someone's profession based on almost nothing, general stereotyping, armchair mental diagnosis, insult based on 'diagnosis' that's needless and honestly irrelevant. Reminds me of the common backhanded insult on reddit, "you must be a blast at parties"
What kind of pilot would be taken seriously if they mix up such a basic aspect of professional knowledge? Jets and prop planes are very different beasts.
I can't think of any job where you can just casually mix up different classes of objects and not eventually have it result in some significant failure.
"Why did you feed the mules and not the horses?" "Don't be so autistic. I bet you're a blast a parties"
"Why did you give me cheeseburgers? The customer ordered plain burgers." "Don't be so autistic. I bet you're a blast at parties"
Yes, being specific about the things you work with is generally a sign that someone is good at their job, but it generally doesn't have much to do with autism. But I guess being condescending is what makes a good cocktail party companion?
While LLMs and people are questionable in their ability for one-shot answers to complex things, with an LLM you can at least ask questions ad-nauseum all the way down the tree, ask for sources, ask it to be self-critical, think step-by-step, etc. From there, you'll at least be armed with more knowledge to ask better questions, whether to an LLM or a person. I think it's also a good exercise in figuring out how to break complex things down into smaller parts, and figuring out what questions to ask -- especially important if it's something where you barely know where to start.
Humans have a tendency to over-value their personal experience and cite their limited knowledge sets, beliefs, and intuitions as fact, and will probably tend to only show you the info that aligns with that.
I guess I'm biased, but for the most part, I don't think the error rates between people and LLMs are significant enough for me to want to deal with the human ego, versus an AI with infinite patience. There are certainly equally intelligent and gracious people, but I don't think they hang out much on Stack Exchange (or much of the popular internet, really)
>This raises the question of whether the emergence of the ability to produce coherent English text only occurs at larger scales (with hundreds of millions of parameters or more) and complex architectures (with many layers of global attention).
>In this work, we introduce TinyStories, a synthetic dataset of short stories that only contain words that a typical 3 to 4-year-olds usually understand, generated by GPT-3.5 and GPT-4. We show that TinyStories can be used to train and evaluate LMs that are much smaller than the state-of-the-art models (below 10 million total parameters), or have much simpler architectures (with only one transformer block), yet still produce fluent and consistent stories with several paragraphs that are diverse and have almost perfect grammar, and demonstrate reasoning capabilities.
The point of TinyStories isn't to serve as an example of a sophisticated model, but rather to show that the emergent ability of producing coherent language can happen at smaller scales, and from a synthetic data set, no less. TinyStories is essentially the language model equivalent of a young child, and it's producing coherent language -- it's not producing grammatically correct nonsense like the famous "colorless green ideas sleep furiously" phrase from Chomsky.
>but I haven't came across many synthetic datasets that are of high quality
I'm not really sure what your personal experience has to do with the viability of synthetic data; it's already been proven to be a useful resource. For example, Meta directly stated this upon the release of their Llama 3 model:
>We found that previous generations of Llama are good at identifying high-quality data, so we used Llama 2 to help build the text-quality classifiers that are powering Llama 3. We also leveraged synthetic data to train in areas such as coding, reasoning, and long context. For example, we used synthetic data to create longer documents to train on.
>Wheel of Fortune hostess Vanna White had established herself as a TV personality, and consequently appeared as a spokesperson for advertisers. Samsung produced a television commercial advertising its VCRs, showing a robot wearing a dress and with other similarities to White standing beside a Wheel of Fortune game board. Samsung, in their own internal documents, called this the "Vanna White ad". White sued Samsung for violations of California Civil Code section 3344, California common law right of publicity, and the federal Lanham Act. The United States District Court for the Southern District of California granted summary judgment against White on all counts, and White appealed.
>The Ninth Circuit reversed the District Court, finding that White had a cause of action based on the value of her image, and that Samsung had appropriated this image. Samsung's assertion that this was a parody was found to be unavailing, as the intent of the ad was not to make fun of White's characteristics, but to sell VCRs.
Maybe it depends on which court will handle the case, but OpenAI's core intent isn't parody, but rather to use someone's likeness as a way to make money.
Scarlett voiced Samantha, an AI in the movie "Her"
Considering the movie's 11 years old, it's surprisingly on-point with depictions of AI/human interactions, relations, and societal acceptance. It does get a bit speculative and imaginative at the end though...
But I imagine that movie did/does spark the imagination of many people, and I guess Sam just couldn't let it go.
In economics there's supposedly something called "induced demand" that also applies to a lot of human behavior in general. Basically, give more resources and capacity to someone, and they'll just fill the space, and not necessarily more efficiently.
AI-assisted answer:
Busy roads, so we build more roads, right? Except no, it just makes more traffic.
Give more time for people to complete a project, and they'll still just end up with a crunch time at the end anyways. "Work expands to fill the time available."
People make more money, then they just spend more money. Buy a bigger house, and it gets filled with more dubious value stuff (conversely, move into a smaller home and realize how much pointless junk you bought) "lifestyle inflation" or "consumption smoothing"
Project behind schedule? Throw more money and people at it, right? But does it help much?
A lot of modern software is arguably suffering from major inefficiency bloat, both in file size and hardware requirements.
So it's probably not quite as "obvious" of a solution to just build more power -- there has to be some incentives to encourage efficient usage instead of just throwing more resources at a problem, otherwise it encourages a long-term build-up of inefficiency.
Here's the other side of that, which is equally as somewhat counter-intuitive:
There's also the matter of the potential inefficiencies of a plant that produces way more than is actually being produced, in which case it's a very expensive waste -- afaik most power plants can't just dial up/down their output to a large degree. Then there are environmental, social, and civic problems, which I guess are easier to bulldoze over in countries that might give less consideration to its citizens. I'm sure there are plenty of other considerations, which you can probably get a good critique on from your favorite AI service.
>Occam's razor is the problem-solving principle that recommends searching for explanations constructed with the smallest possible set of elements
it's a general principle or _recommendation_, not some law of the universe. Some people might argue that "God wills it" is the simplest explanation for many things, but that doesn't mean it's true. The simplicity of an explanation doesn't necessarily have anything to do with its validity.
In addition, the introduction of panpsychism, just like the introduction of God into any argument, brings up a whole other set of questions that need to be answered -- additional complexity, which is the opposite of Occam's Razor.
Emergence out of complex systems is arguably the simpler explanation, because it's something that's already been observed, measured, and studied, like storms emerging from simpler principles of weather systems.
Or take your computer or smartphone -- do you truly understand all the mechanisms from which we go from "shocking rocks" to create series of on/off signals, to things like communicating on the internet, or watching videos? Is computing and mathematics some inherent property of silicon? Nearly part of a computer, on some fundamental level, is a relatively simple mechanism, and has an almost useless function on its own. Even for engineers who understand every level of abstraction, it must still be near-miraculous that any of this works, even though these emergent properties are deliberately crafted, well-documented, and understood.
His ability to not only understand, but to also effectively communicate these concepts, I'd say makes him one of the smarter people out there. And yet, he remarks, "I don't know about you, but it really doesn't feel like this should actually work." There are still things people don't understand about why AI works the way it does, despite the fact that we built and trained them -- feel free to hit up Claude or your favorite resource for examples on emergent properties. LLMs can be passably apt at things they weren't trained for, and exhibit behaviors weirdly similar to people (like confabulation), despite the fact that their exposure to the world is literally only text.
I'm already imagining ways people could twist this into proof of panpsychism. But the point I'm getting to is that the human body is an absurdly, stupidly complex system of 37 trillion cells. The Milky Way is estimated to have 400 billion stars, at most. Like LLMs, we understand some things about our brains... but the complex interaction of many parts is less easy to understand. The purpose and value of feeling and awareness as a function of survival isn't a "tall order" -- it's just difficult for the human brain to grasp so many moving parts simultaneously. For some people, the complexity of the eyeball alone is proof that there must be a god -- the sheer magnitude of billions of years of brute force trial-and-error is difficult to comprehend.
Human intuition: a potentially powerful, but simultaneously and often error-prone weak force of the human brain.
>Panpsychism requires that the universe updates its state by conscious choice, which we already know happens
[citation overdue]
I think there are are at least two levels of logical fallacy here, not to mention avenues of undefined and fuzzy circular logic, but I've already spent too much time on this. I'd say try pasting that into Claude or another "big AI" and see what their critique is.
I think most custom fine-tunes and merges on HuggingFace will do this unless they specifically mention it being censored. Even the lower param models have been surprisingly good, with relatively fast progress being made in the 7b and 11b models.
My "daily driver" is Fimbulvetr v2 11b, surprisingly slapped together by an EMT. Kunoichi 7b seems to be a pretty popular model too. These can be run locally with as little as 8 GB free RAM (preferably VRAM) with an easy install solution like LMStudio or Faraday.
You can generally find a lot of recommendations in places like SillyTavernAI or LocalLLaMa on reddit:
>obviously creates predators as more people are exposed. (from your earlier comment)
>There isn't a single person in the world
Is there any good research to back these up? Societies have had similar lines of reasoning involving prohibition and banning in the past that seemed "obviously" intuitively correct, some of which have been mentioned already by others.
As mentioned by someone else, CP is "radioactive," and as such, in a black market would probably fetch an appropriate price. It's been an extensively studied and observable behavior that people can, and will partake in unethical or immoral activities if the economic incentive is high enough. It's not an either-or thing; it can be about the abuse, but it can also be about money, or even just a lack of empathy. The "lesser" crime of CP distribution doesn't necessarily have to be about the abuse either; people can find themselves in crime rings out of desperation, or coercion.
If realistic AI-gen CP can greatly devalue the real stuff, the risk would become much less worth it, except for the types of people you specifically mention. As I mentioned in another comment, it's possible to train and generate victimless synthetic CP -- AI learns concepts, similar to how a human artist might piece together concepts to create something they've never seen before.
The idea that virtual content can't provide some sort of catharsis or outlet for at least some people is also questionable, unless we can get a significant number of actual pedophiles to come and testify to this. People already find outlets in acting out any number of emotional, sex, or abuse fantasies via text LLMs, despite it being only text. And generative AI is becoming increasingly realistic, including audio and video domains, even if it's disturbing to think about.
There is evidence to support that pedophilia is to some extent a result of biology and uncontrollable life circumstance, and there are non-offending pedophile support groups, so avoiding abuse crimes is a choice for a significant non-zero portion of pedophiles.
>The major problem with CP is that the most cost-effective way of producing it is abusing children.
I'm not sure that's completely true. Despite the surprisingly widespread notion that generative AI can only strictly reproduce things it's been trained on, it can still deductively create novel things based on that data. For example, image generation models are trained on a wide variety of adult human figures. Children are just... kind of a subset of what makes an adult, so via some clever prompting, it should be possible to steer a model to generate CP. It should then be possible to generate synthetic data to better tag, fine-tune, train a new model, or whatever.
The problem with this is a person needs to steer that ship -- who's going to do it? Even if it has the potential to prevent abuse, it's something that's so universally unacceptable that people generally don't want to touch the topic. There are law enforcement officers whose job it is to manually search for and assess CP on the web, but given the underlying philosophy of law enforcement, I'm sure most of them would not approve of the idea of generating CP, even if it's synthetic. So... the ideal pick would be an actual (hopefully non-offending) pedophile, but who wants to be known as an employer of pedophiles?
There are supposedly communities of pedophiles who wish to stay non-offending, and I imagine it'll be an "outsider" group that tries this, if at all.
For example, when I first heard a deaf person laughing or talking, I probably internally noticed how... different the sound was. I'm guessing most hearing-abled people have a similar experience. It's very... unfiltered? It made me wonder how much even my own laugh was sculpted by my environment. If I relax my voice, I notice my voice becoming much more booming and obnoxious than my "normal" speaking voice.
Anecdotally, I've noticed Japanese people are much more likely to have a sort of stifled, raspy, restrained laugh, even when they're in a situation I might expect an American to have a belly laugh.
A lot of cultural values are encoded in language too, and in turn, the languages we speak can affect how we think or interact. Anecdotally, my personality is a bit different depending on what language I speak. I think the concept of what's actually encoded in language is being explored with regards to how/why LLMs "feel" smarter than they really ought to, or seem to show intelligence beyond simple "stochastic parroting"
Somewhat more obviously, most people will "code switch" not only their language, but vocal tone or even demeanor, depending on our current "persona" or our audience. Recently, Paris Hilton entertainingly demonstrated this: https://www.youtube.com/shorts/g9pal1ConNU
And this is more of a half-baked personal speculation based on a scattering of theories and case studies, but the environments we live in, the narratives we expose ourselves to, the people we surround ourselves with all probably very heavily define a lot of our values, beliefs, and even personal preferences, to an extent that would disturb a lot of people. Self-serving biases, post-hoc justifications, and confabulations give us convenient and creative ways to validate our own free will, volition, and independence, but I often wonder how much of our supposedly great human intelligence is an uncomfortably thin veneer on a largely automatic pattern-absorbing sponge.