LLMs are writing as well as humans for people who think writing words is the goal. Printing presses also write words better than humans. LLMs are solving the problem one layer higher than printing presses, but still several layers down from the top.
Note that LLMs aren't useless, since they have other strengths which can make them equal or better than humans in certain applications. For instance, they can tirelessly review code for silly mistakes, making them useful in cybersecurity.
Finally, someone else understands that writing is not the goal, the entire point of all of this is our communications between one another and this market for narrative stories is a nice departure and it does convey important knowledge and moral lessons that larger society needs. But it is an indirection, an indirection to the mission of communicating solutions to how to live as a human and that is why we care, why the interest persists despite lifetimes of narrative dreck. We are adrift without instructions, and no religions are not the instructions.
I think the first goal of writing is to talk to yourself. Without an external, unchanging scratchpad, most people cannot keep their thoughts truly straight. It can feel otherwise, but trying to write something about any tricky topic even a few pages in length can take a long time to be self-consistent. Clarifying it takes even longer, especially when you intend to share it. Writing as thinking is definitely a thing. I believe it’s the trickiest part of understanding how such interesting things pop out of a language model trained to predict the next token and rewarded with pleasing sequences. Since it is not thinking, it proves that certain sentences (most?) can be generated thoughtlessly.
I say that LLM's are a pacifier. LLM output is to conscious output as a silicone nipple is to a breast.
Sure, at some level, there are some people who at least some times, cannot pecieve a difference. That does not prove that there is no difference or that the difference doesn't matter.
I think that LLM's are a broken mirror or maybe disco ball. What you get back is almost always a reflection of what you put in, just garbled through and with the reflections of so many other people. The best parts that I often get back are mostly just the stuff that I already put in it.
I think that's also where LLMs get the most dangerous. That ELIZA effect of telling you exactly what you told it but in a slightly rephrased way makes you think that you think that your ideas are genius. You've got someone "actively listening" and telling you whatever idea you just gave it was the best idea it ever heard.
That's not what you want from a serious writing partner. You want a writing partner to challenge you, to point out your mistakes and flaws, to make you better, to offer a truly different perspective to your own. You don't want a dark reflection from a "yes man" to spiral into.
I think it is a net positive for me, because I am super unconfident by nature and the agreeable, almost sycophantic output from major LLMs give me just enough of a boost to debias me. They are certainly better than a physical rubber duck.
I feel it is a net negative for me. One of my hobbies has always been fiction writing. The LLM reflections of my own writing become slot machines to me for cheap dopamine hits. It's not because LLMs are good at fiction writing, far from it, it's that LLMs are good at reflecting my own writing back to me in fun house mirror ways that amuse me and frustrate me and pleasantly surprise me.
That slot machine addiction gives me a lot of perspective on what LLMs are good at in coding versus what they appear to be good at. It's easy to characterize them as "over-eager junior developers" or "powerful rubber duck debuggers" but their biases are still your biases so long as you are the one writing the prompts. A lot of the "surprises" aren't intelligence but fun house mirror versions of the same biases you input into the system.
I keep referring to their coding capabilities as "legacy code as a service" and the addiction to the fictional equivalent only underscores that understanding of LLMs. You don't write the next great American novel getting a word salad of past great novels back full of the best and worst of your own writing. You don't write the next great American software getting a legacy code jumble of past great software back full of the best and worst of your own prompting.
At the end of the day I'm still a pragmatist and using the LLMs in my programming efforts, but I don't trust them as a partner, I use them as a tool. (A biased tool that I know needs a careful eye.)
> but their biases are still your biases so long as you are the one writing the prompts.
That is if you only work with your LLMs / Agents in a "straight ahead" manner, versus using knowledge of the bias baked into AIs and purposefully invoking a specific type of subject matter expert for problems that require specialized knowledge. It is similar to using people skilled at what you need, invoke through the vocabulary and language of your prompts specialized skills that are not tapped until using the terminology only those specialists use, and like magic their skills are suddenly in-context and the AI acts accordingly.
Don't just focus on the model, half the turns in a session belong to the human, it's an instrument that depends on how well the user wields it. It does not need to produce the novel insights, just to support the user to have them.
I think narrative stories are a "brain hack", in the same way heavily processed food, drugs, clickbait, soap operas and TikTok are brain hacks, combined with some exclusivity status signaling games. They're just old enough to have gained a cultural cachet that more modern forms of entertainment don't yet have. Same for art, music, comedy and culture generally.
There is nothing wrong with those "brain hacks", if they make us happy and don't have negative consequences, why not use them, but we should recognize that it's what they are.
I think that an alien intelligence could communicate with us about Ohm's law and Rieman's last theorem, maybe even debate resource and power allocation strategies (democracy, monarchy, capitalism, socialism) with us, they could probably understand the pull of sports, even if they couldn't participate in it themselves, but art and humor would be completely lost on them and considered a weird human excentricity.
This sounds like a fully general argument against being alive and having experiences; you could equally well call love a "brain hack". I also care a lot about solving practical problems, but the reason it's important to solve them is because it leads to human flourishing, which includes participating in all these activities you're deriding. Otherwise why are we doing anything at all?
The purpose is paying taxes, interest, rent and various tributes to the full extent of your capabilities and to sacrifice your life for the benefit of your rulers if ordered to.
The "story form" is a brain hack, i.e. a way of disseminating "problem X may cause Y and be resolved by Z". Whether that's useful depends entirely on what's being communicated and how.
Art, music, humour, poetry are more basic and the commonality is more or less "I bring you surprising phenomena". If phenomena are surprising to you then your mental model of the world needs improving.
Culture and art can teach something to its audience, unlike most of the garbage you likened it to. It's clear you don't hold it in much regard, since you compare it to outright scams (clickbait) and even imply that consuming fiction is basically done for virtue signalling.
But like it or not, humans are not computers. They are animals with messy, complex thinking and reasoning that is impossible to fully describe, and incredibly complex societal systems created from their chaotic interactions. These messy people don't learn from datasheets and reference manuals, even if you'd like it to be otherwise.
Of course, there's a place for non-fiction, I like some of it a lot. But art has subtlety to it - instead of smacking people in the face with data and lists of objective measurements, it can pry into them by using the human ability to imagine themselves in other shoes, experience emotions that aren't results of real life events, and much more. There is a reason why parents teach their children values by reading them fables or educational stories, not the legal code of their jurisdiction. And this is also why fiction actually changes the minds of people reading it, arguably doing it more effectively than nonfiction. Slop (clickbait, average tiktok content, whatever else you're imagining) doesn't make you feel new things or think of something you haven't thought of before, art does.
The aiens are also a bit of a copout. You're just assuming that there's bound to be this perfect species (that are effectively computers) that operate only on mathematics and discrete logic, who are obviously superior to us and can show us our wrong ways. What if computers aren't always superior? Or what if organic life requires some of that messiness you don't like to be viable? What if the aliens had their own weird things we could never relate to?
I understand where you are coming from, that brain hack is how we remember cause and effect. It's very much like the use of Jazz Hands to signal to the dullards "now is the moment you're supposed to remember and say 'oh, I liked that'" A narrative story used as a vehicle to convey moral lessons and life guidance is the Jazz Hands of our fragmented and incomplete human instruction manuals.
What's the purpose of this? You didn't disagree, you just went on a tangent. Is this another "brain hack?" If so, there is something wrong with them. They often make words appear to be something that they are not.
Like a magician would use a thing called "indirection."
We already suspect that many non-human animals have a "sense of beauty" and a sense of humor. At least one other species spontaneously dances on the "beat" to music (parrots).
I think alien intelligence would be able to appreciate art and humor, even without necessarily feeling it viscerally, as we do. We know that our particular brands of them rely on highly-developed and specific aspects of our psychology, but at a basic level, they seem to fundamentally be about congruence, incongruence, expectation, and interpretation/perspective. Any embodied intelligence that can observe physical reality is capable of grokking those concepts.
You're both missing the point, I think. You're missing that AI and the people who utilize it in secret will fool readers. Today, you cannot watch a video of something amazing, submitted by a stranger, and trust that it is real. People who don't doubt the source of a book will read AI-created content thinking it is the real thing.
Reputation will mean more than ever before in the age of AI. People doubt "mainstream" news, but their reputation for verifying sources and stories for truth will be the only way people can trust what they read and see. Novelists and creators will have to build trust with their audience, if the audience is caring enough to demand authenticity.
I'd like to live in a world where people only make art to serve a higher purpose. But we don't live in that world and never did. Before we had slop we had pulp — formulaic, low effort writing full of cliches of romance, lust, crime and adventure broke college students could churn out en masse to make a living.
I'd also recommend having a look at the books published by this article's author. I will admit that I have never read nor heard of any of them. But come on, just look at the books published covers and the way the author introduced them and tell me that they are not industrial products engineered for mass commercial appeal.
We do live in that world, but that world is also filled and conditioned by the predator and prey structure we all operate within. The higher purpose that Art serves today is aliased, hidden, and conveyed in layers of metaphor. This is somewhat of a game, as the predators playing the art game try to consume their prey, anyone that is not an art philosopher in the pursuit of stories to explain our situations and not fame or wealth. The obvious paths were long ago traveled and farmed to barren land, and all that is left is the extremes as nobody wants to admit that this, this with all of us, is it and nothing more.
Many of us understood that, and tried to tell it over and over. My keyboard is worn, my mind is numb, my soul is weary from typing/telling that the output is not the aim, but the content it carries, the substance it holds.
Yet a very vocal group advocates integrating LLMs into every aspect of their lives, and the other group watch in dismay.
On the other hand, I used a sentence form used by LLMs in a comment during a morning-brain moment, and an angry mob tried to crucify me with torches and pitchforks.
Many people love to generate text with LLMs, but most of same crowd apparently hates to read LLM output. An interesting phenomenon.
It must be a distinction between engineer-type brains and leader-type brains. If you're focused on social status climbing, the fact that you produced a report, and the length of that report, is far more important than whether the contents of the report have any relation to reality beyond the surface level. This is true if you're trying to get VC funding as well. It's even in institutional academia, as seen by the Sokal affair.
I thought for a second you were going the opposite way with the brain thing. Good leaders are very much concerned with accurate and timely communication.
It can be a very engineer thing to point at a vanilla Doxygen dump and call that thing "documentation".
It is documentation. This week, I looked at the Laravel API docs[0], just as much as I looked at the framework guide[1]. It all depends on the question you have and the answer you need.
It's like the car manual (Do they still ship with one?) where every detail is laid out. It's not important for daily driving, but when you to fix an electrical problem, that's the reference you need.
It is documentation, but unless someone writes it, there's nothing "meaty" there explaining why certain choices were made, which APIs which will be deprecated, the overall design, where new things should go depending on what they do etc etc.
> is far more important than whether the contents of the report have any relation to reality beyond the surface level.
This only works in broken companies where leadership is already overrun with other status climbers.
If someone tries this trick at a functional company and is discovered, their reputation takes a huge hit that can take years to overcome, if ever.
At high performing companies reputation is everything. It can be a real problem to even get people to work on difficult or risky problems because they’re afraid of having a failure attached to their reputation.
Throwing unreviewed LLM slop around would be a reputation death sentence if your peers are good at what they do, because by now we all know that it’s easy to generate long reports with LLMs. The secret is out and anyone who follows technology knows it.
There are some companies where you can LLM report-slop your way to prestige. Those are not companies you want to stay at.
The promotion of LLMs is, I believe, basically a referendum on how many companies are broken in that way. So far, the adoption statistics are looking like it’s most of them.
To be frank, I neither live/work in an environment or live for "climbing" anything, so fluffware is not something I produce or understand.
On the other hand, I have friends who play that game, and I don't talk with most of them, because their life views and mental states continually drift to a state which makes no sense whatsoever for me, so we grow apart.
You are describing the bifurcation of society that is the moral lesson of The Time Machine by H.G. Wells. The Morlocks are the scientific and engineering "gender" (so to speak) of the human race while the Eloki are the child-like humans kept around because they are the food the Morlocks consume to survive in that future.
Yes, I've read the book, the two movies (one modern one from the 60s?), and the sequel, very well done!, by Stephen Baxter.
It's a good concept, but for some reason I thought of a military movie I watched recently, where someone was going on about how they were third generation military. And how their friends and family were military. Sort of the same kind of siloing there.
Ilium by Dan Simmons was an exceptionally good read, if you're looking for another review. Often people just haven't heard of him, even well and widely read people, though he has several great series.
I don't read info on books ahead of time, and sometimes don't even read the summaries. I've read a LOT of scifi, must be in the thousands of novels by now, but I was completely unsure what was happening for the first 4 chapters at least, and in a very, very good way.
> but the content it carries, the substance it holds.
It's even more than that, I think. It's the intent to reach out and transfer to others something that you experience internally. It's a sense of purpose motivating that content and substance to be packaged up and communicated in the first place.
Sometimes LLMs will slip up and say things like, "I see this working with people all the time." And it REALLY breaks the illusion that the text they extrude means anything. Because obviously there is no "I" and no, the LLM does not have this autobiographical memory the text is simulating.
My wife uses ChatGPT a lot for advice on child development and parenting in general. Every so often, it says things like "in my experience, what I've noticed works best is..."
> Many people love to generate text with LLMs, but most of same crowd apparently hates to read LLM output. An interesting phenomenon.
Are you sure it's the same people? I feel the opposite: it's the same people who happily read tons of LLM output themselves and send some of it to others.
my coworker presented a 93 page pdf for a project plan entirely created by $230 of Claude's deep learning. i hope i wasnt too mean while asking him variations of "what now? what do i do with this?" the answer is obvious: read it. but so much of it covers problems which are already solved and my coworker clearly hasnt considered at all
point is: maybe llm output is best not shared. maybe, like a dream, the only audience who can appreciate the content of the writing is the first party
Having made the mistake of writing a couple of them, I’m pretty sure the only uses of non-fiction documents longer than ~30 pages is
* As a reference; people can look look up just the algorithm they want
* As a token that says “I’ve thought about this problem”
(The former is somewhat a subset of the latter. All of the stuff they didn’t read when they used the work as reference was there to provide them confidence that the algorithm they plucked out of the thing had some thought behind it).
totally. ive made some 10 pagers documenting topics ive THOROUGHLY considered and had my work peers review. same as you, the purpose is to point at it 5 years from now and say "the answers in there"
but when my coworker pays to create documents on topics he hasnt considered, for ideas of projects which dont exist, it feels like an exercise in text generation and i become mildy put off for the onus of reading it. i know it was written without purpose and while there are possibly novel tech and ideas in there, it feels like being handed less than a rough draft because it lacks any aim and intention
now, i expect he would get much more from it. hes the first party audience who reads the doc and sees a valuable project somewhere inside of it
Sometimes the medium is the message though. Especially in literature, prose matters. It's not just about what is said but also how.
This applies even in "low" art. AI generated imagery from mainstream models tends to be so ugly it detracts from effectively communicating what the prompter intended.
No one likes generating content. Many like generating income or status. It is a new era. The era of curation. I would predict shorter and weirder stories and meme style communication patterns outside normal syntax. Poetry e.g. is replicable but not redoable. I am ambitious we are to see the best writing ever soon.
- Since I can generate code, writing it feels mundane now. I don't write code anymore.
And they continued how they replace almost everything they do with LLMs.
I have seen people who never think but spend all their days querying LLMs for everything. Docs, help, howtos, system administration, coding...
What I see is an incentive for less work, less effort. Not income or status. They don't spend their "saved" time, if there's any, for personal activities or self-improvement either. They just do things with less effort, and just live as they lived before.
It's brain numbing.
> I would predict shorter and weirder stories and meme style communication patterns outside normal syntax.
I don't share the same sentiment, yet I don't have idea what to expect. It has to get darker to get brighter again.
It is very weird. 6 months ago I generated some documentation for a service that was largely undocumented. When it was done my only thought was "who is this for?" Is any one going to actually read this or are they just going to ask the LLM? If someone didnt write it, why would I read it? It just becomes something to pipe into the context for the next person. It seems though a lot of people don't have this experience, its like they think they are the only ones with access to this technology.
Additionally I am so, so ,so sick of reading claudes overly wordy technical writing ish bullshit. Its exhausting trying to find the fuck ups or hidden assumptions in the pages of text it produces for everything.
Every time I try and search for ways to make it stop I just find people trying to save tokens which I don't give a shit about. I just want it to communicate things to me without having to constantly read 4 paragraphs, dump the context or fork it or start a new session so it "remembers" or re-weights that is was told to be concise and terse in like 7 different places. Maybe I am using it wrong.
For my medical notes assistant I had to set a global "write in a telegraphic, minimal style" because it was fluffing up every aspect of the notes. Nobody wants to read that BS and it's impossible to find information in a series of walls of text.
It can also go the other way. It outputs a paragraph so full of telegraphic technobabble that it’s just unreadable, inventing new jargon every other sentence, using terminology invented in its thinking stream and never explained. And when it combines the two issues and it outputs 4 pages of technobabble, it gets just exhausting to read. Paradoxically, given that 2 years ago everyone was using LLMs to summarize web pages, their ability to condense text (including their own thought stream) is still lackluster.
I dream of some post-processing diffusion optimizer just condensing all of their slop away. Until then, I just tell it to _avoid_ telegraphic and newspaper-title speech and to output one or two paragraphs of prose only.
To be fair, LLMs didn't really change this that much: Throughout my career, I've been generating docs that nobody is going to read, only because some process somewhere says "In order to get engineering approval to commence work, an Engineering Risks Document (ERD) must be produced." Everyone knows that nobody reads these, but someone with a rubber stamp checks to see that the doc exists and it looks sufficiently long, and then he stamps the stamp. LLMs just make producing this shit faster.
They are quoting John 1:14. At the beginning of his book, John describes Jesus as being the personified word of God. That was likely an early example of the influence of Neo Platonism on Christianity.
In the beginning were the Words, and the Words made the world. I am the Words—the Words are everything. Where the Words end, the world ends. You cannot go forward in the absence of space. Repeat.
Wow. My comment is auto-collapsed. I'm not aware of this phenomenon on HN. Is it a moderation thing? It hasn't been downvoted. I don't think I've been shadow-banned or had this comment flagged or anything.
Anyway, to answer your question, I was pointing out that the request for "important knowledge and moral lessons that larger society needs" and "communicating solutions to how to live as a human" as well as lamenting that "We are adrift without instructions" is actually a request for religion. Religion is how to make sense of how to live. So the comment is asking for religion and rejecting religion at the same time.
At the same time, I was pointing out that one religion in particular has directly addressed the point that "writing is not the goal." The "word"—or "logos" in the greek—is the reason, the logic, the rational principle of the cosmos behind the words. It points to that need for a solution for how to live, as well as showing there is a real answer.
Yo, this blog post could’ve been a freakin’ prompt. I have years and years of blogging for an AI to learn from, and over 20 novels, and tons of short stories. One prompt, and all this spills out. And if your goal is a blog post, you’re gonna love this era. If your goal is the joy of thinking in words, nothing has changed. So this is the real part, right here: Doing a thing for the love of the thing is no different than it ever was. The problem is, there aren’t many people for whom this applies.
You can also view them as data expansion tools. They expand an input to an arbitrary long text. That text contains roughly the same amount of information as was contained in the input to begin with.
This is not very useful unless your work is content production. More useful is when the output length is shorter than the input. Then it is a (lossy) data compression tool. These are, as we all know, very useful for images and sound but now we have them for text too.
This assumes one-shot generation. I don't use it that way, and I don't think many people who find it useful do.
I use it as a colleague I'm drafting with. I'll ask how a sentence lands, tell it to make a paragraph more friendly or more dramatic, have it argue against a point I'm making. The piece comes together over many turns, and the information arriving in each turn is mine — it's not expanding one prompt, it's helping me converge on something.
The information-content framing also proves too much: a copy editor adds no information either. Neither does a translator, or a ghostwriter. What they add is form, and form is most of what makes writing good or bad.
Many people make the mistake to think LLMs are stochastic parrots, but they are more like pianos, you play the keys and they make music, if you don't bring no new spices to the soup no wonder it comes out bland.
I really hate the whole thing about companies promoting using LLMs to change one-liner emails into this long multi-paragraph thing that ends up saying almost nothing ... just wasting everyone's time in reading them (with an extra sprinkling of hallucinations).
LLMs have created more billable hours for me as an attorney because the LLMs convince the client that they need to know more about irrelevant issues, how specific statutes may apply, and other such things. They may ask about specific procedural issues that they would otherwise have no idea about that I then need to explain in greater detail. It reliably bloats things that would be a .1 into .5 or more.
>They may ask about specific procedural issues that they would otherwise have no idea about that I then need to explain in greater detail.
My favorite part about LLMs is that their insufficiency as legal aids (Aides? Aids.) seems to be slowly cluing-in the general public as to how Byzantine and "inside baseball" our legal system has become. (This is a bad thing.)
It's a big part of why I'm strongly opposed to common-law systems (as opposed to civil law). All the case law, all of it becomes, well, law. And then you get the whole "legislating from the bench" critique, which, while valid critique, is only a big problem because it's a common-law system.
Yes, I realize many (most?) legal systems are actually hybrid, but they do tend to err pretty strongly into one direction or another.
I agree that the output is more likely to be valuable when it’s shorter than the input. And when they use tools to do research and try things, it can be a lot of input that comes from all sorts of places.
I use them as text consumption machines. They read and watch a lot of the books and movies i have no time to. And from the TLDR I decide whether the full thing is worth watching.
That is not how it works because of how bloated everything is. There is rarely a movie, book or series lately that are not 2 times longer than needed. So you are forced to go for aggressive recaps.
Most of the non pulp books published - contain at least two times as many pages as needed for the story or narrative.
The information density in the modern world is absurdly low.
Yes, I agree on modern trends for that. There are some modern authors not too bad in that respect, but it may be why I find older sci-fi books more appealing.
LLMs don't do anything well by human standards. Not even the one thing (programming) they are supposed to be good at. It's disheartening how many people are prepared to accept the slop an LLM puts out just because it's fast.
Possibly, although I have to say that I get PRs now where it's clear it's a miss by the LLM and was generated by a human who did not take the time to understand the subject matter.
In the past, a human producing humanslop would get a serious talking to, but nowadays it's apparently a valid excuse "well, an LLM generated it". This approach is placing a huge burden on the reviewers, much more so than in the past.
The difference with 'humanslop' is that it's avoidable by training or hiring a better human. With AI it's all the same, even SOTA models are still very distinctly and obviously AI in what they produce. And even very dedicated humans have a limit and aren't capable of creating low value content nonstop, meanwhile you can set up an LLM, point it at the internet and let it start filling gaps by firehosing it in low quality slosh 24/7, completely unattended.
The interesting thing for me is that I have a very easy time of identifying AIslop. Have I been conditioned to simply ignore or accept humanslop? Is that the direction this is all going? That at some point we will simply accept both as a product of the human existence?
I really want to know what people specifically mean when they allude to some "higher" level of cognition that humans have above LLM's when it comes to useful outputs. I don't mean the nature of consciousness or internal experience (which is unfalsifiable), I mean the essence of cognitive labor as a tool: writing is about relating concepts, drawing comparisons, aggregating elements of reality into a fictive but believable narrative.
All these processes can be reduced to mathematics and computation. The human brain is just a tool, like any other organ. Do we deny this reality because it shields us from existential horror?
> I don't mean the nature of consciousness or internal experience (which is unfalsifiable)
The way I've been thinking about it is not that internal experience is unfalsifiable, along with the modern science connotation that everything that is unfalsifiable is automatically irrelevant (which would be quite absurd, as everything that is dear and meaningful to humans are all simply cognitive phenomena in the end!). It's that it is externally, naively and a-priori unfalsifiable. That is, you can't externally measure or prove it, without having an a-priori theory of what it is and how it happens in the first place. But it is internally falsifiable, in the sense that if we have it, then we know we have it. Note that it's not that "If I have it, then I say I have it", or that "I one has it, he says he has it"; those are all communication to the external world, and for example machines can clearly say they have it without having it, or you can have it before even learning the words that represent it. But if you have it, you can know you do. This actually creates some potential for expanding this knowledge, maybe, in the future, to allow for external verification. We can internally investigate properties associated with consciousness, and from those properties (along with scientific knowledge), maybe one day we will know more about the objective basis which affords consciousness/sentience manifesting or emerging. So I see it like an experimental field where the component of self-assessment or self-experimentation or self-measurement is essential (again along with more conventional scientific notions).
I have been thinking the same way about all properties of life/consciousness, in particular if we think of ethics as a property of (even if you consider the ethics or ideals only pertaining to how it would be ideal to lead your own life) the mind, or of sentience in general. I think of ethics as an inherent property of existence: it stems from the fact that internal experiences/internal life/sentience/etc. are the only thing that could have value, that in fact sentience is experimentally (again taking the point of view of experimental sentience) the only thing that truly matters or has value. If we are minds, we know that only the content of minds can matter; it cannot matter, for example, that a lone rock in some place no one can know about has this or that fate, because there is no mind to interact or perceive and be affected in any way by it. So all of ethics can probably be founded or based upon this kind of observation, and the act of experimental self-observation of consciousness. And I think it's quite important that we collectively recognize those values and what matters to make our lives better (better yet, as good as it can be!) :)
Going further, all of those are currently under the umbrella of metaphysics, which is again a kind of prejudiced-against (often taken as irrelevant, when it contains some of, and in fact many of, the most important things in life) field of science/knowledge. I think metaphysics should just be seen for what it is as a more indirect kind of science. In a way all of science is varying degrees of indirect, we use all sorts of instruments and need to make all kinds of assumptions (that conceivably, but unlikely, could indeed be false) to be able to do any kind of science or assume any kind of validity to results. We need to assume (although indeed many of those assumptions we don't even need to consider, i.e. they're implicit) there aren't any "gremlins" of whatever kind interfering with our measurements. We do our best to cross-verify whatever experiments we have to increase our confidence. Likewise, metaphysics to me is simply about inferring facts and properties with more indirect methods. One of them is what I referred to above, i.e. observations from within your own mind, and proposing properties of cognition from first hand experience. A second (very close to actual physics) is conjecturing what other universes are out there of if there is some larger mechanism our own universe (or all universes) are part of. Another is judging other properties of existence, for example, are our minds completely separate or are we all ultimately part of the "same thing", somehow? How did I come about living as the person I am currently living, am I perhaps a random sample, with some particular distribution, of all living beings? How do those facts influences how we should live together? (I think they all strongly support the notion of valuing everyone, and discredit any absolute notion of primacy of yourself in ethics, i.e. egotism).
The particular problem I have with your take on value is it sets up a future conflict based upon the state that a human has. It's based on a number of presumptions that only a particular set of arrangements of matter, well, matter. And those that matter happen to be watery sacks of meat.
So in your framework you setup a definition of what an external measurement of someones sentience state... Now, someone makes a machine that outputs those states and externally behaves as if it has sentience. What do you then? Upturn the table and get mad because your classification system no longer just contains humans?
Thank you for your reply. I'm not sure I understand your objection perfectly.
> it sets up a future conflict based upon the state that a human has
It's not clear to me it does.
> It's based on a number of presumptions that only a particular set of arrangements of matter, well, matter.
Well, this premise must be true, I think. How could it be otherwise?
> And those that matter happen to be watery sacks of meat.
This I disagree with. My comment contains no such claims that only humans matter.
> So in your framework you setup a definition of what an external measurement of someones sentience state... Now, someone makes a machine that outputs those states and externally behaves as if it has sentience.
So, this is something we need to be very cautious about, I think. First, we don't really know what makes someone sentient. I do not claim to know, and certainly not know with a high degree of certainty (I may have a few incomplete hypothesis).
First, it's clear to me that someone's external behavior does not dictate sentience (positively or negatively). I gave some examples, but I think this can be pretty reliably established. To make things simpler, divide the candidate sentient systems by their "output quantity/quality", from less to more. (1) On the "no output side", it's clear that someone can think a lot and have a wonderful inner life without being able to speak. There are (some medical) cases of people going long periods without speaking by choice or by medical condition. Those people can be sentient just fine, obviously you shouldn't need to speak or "output" anything to be considered having an inner life. (2) On the "human-like output" (or superhuman), the argument becomes more involved. First, clearly just the output itself being indistinguishable from human is not sufficient because it could be for example a simple program that just pastes some verbatim copy of a never seen dialogue. Or very simple chatbots that are quite convincing at appearing human (can fool many judges for a while), but clearly also are not sentient. Even for a for a system that genuinely resembles a real human even with under scrutiny, (with all sorts of similar abilities across all fields), consider the case that this system could be simply one astronomically large (unrealistic, I know) lookup table. Basically, it simply works inside by fetching an indexed piece of data and returns it. This is possible, although this machine would need to be astronomical in scale (so it's more of a thought experiment). It shows to me, convincingly, the actual "internal" or "microscale" even implementation of your thoughts matter critically for sentience. I presume, for example, that even if some small portion of your mental activity (i.e. a kind of "subcircuit" in your brain) were replaced by something like a lookup table, you would lose part of the experience of that mental activity. This is clear when the content of your mind differs when you're trying a (kind of) problem for the first time and when you're solving one for the umpteenth time. The latter (possibly, not necessarily) feels more boring, also normally more automatic with less thinking required as you've built habitual responses.
> What do you then? Upturn the table and get mad because your classification system no longer just contains humans?
I would do no such thing. I would welcome any other kinds of sentience. But it would be important to me to carefully establish and understand what kind of systems are sentient, indeed in large part to protect them.
---
My main point is that you kind of need to start with sentient minds to understand what sentience is (and of course, not only that, but sentient minds that are aware of their sentience and willing/keen to study it). If you start without sentience, it doesn't seem like this particular concept would form a necessary basis for a theory of action or motivation. Like you can't prove from elementary mathematical premises that sentience is important, I don't think. We need to start from some knowledge that derives from first-hand experience of being an actual sentient being, and expand, solidify or formalize this knowledge to be able to understand it with scientific precision (or anything close to it).
>some knowledge that derives from first-hand experience of being an actual sentient being
This is where things kind of fall apart. We all agree (hopefully) that humans are sentient. The question is, "what else is sentient". Are all animals sentient? If so that leaves a very, very wide range on what sentience is. If not you have to operate your scientific scalpel with a very high level of subjectivity on what is and isn't counted. At some point in the past there was no sentience, now there is. Was it a click and we have it, or was it a slow awakening to the world bit by bit (this is my take). It would also quite likely miss systems that could operate at a higher level of sentience than humans, for example some types of hive minds.
My take on it is that any sufficiently complex system with memory, the ability to perceive outside information, difference that information from what it's doing versus what's happening outside of its actions, and learn from that behavior is at least somewhat sentiment.
This leads me to think that LLMs are 'kind of' proto-sentient. They are not persistent, their short term memory is tiny, and they don't self learn. Once we get the persistent and self learning parts working this really starts to check a lot of the boxes that we consider sentient. Of course by the time we get that working it seems likely we'll have much bigger problems in dealing with AI and the entire sentience thing might not be the first thing on our plates.
> This is where things kind of fall apart. We all agree (hopefully) that humans are sentient. The question is, "what else is sentient". Are all animals sentient? If so that leaves a very, very wide range on what sentience is. If not you have to operate your scientific scalpel with a very high level of subjectivity on what is and isn't counted. At some point in the past there was no sentience, now there is. Was it a click and we have it, or was it a slow awakening to the world bit by bit (this is my take).
I believe pretty much the same thing. It probably isn't all or nothing.
But for me personally I'm really interested not only in what it is (although I'm very much interested in that), but because the problem is so thorny (as you mention subjective), I am first centrally interested in how (or if) we can actually know it reliably. So subjectivity is part of reality, as the mind is part of reality. In principle, it seems we should be able to understand quite a bit about subjectivity because of this, because it is part of reality. For example, the mind clearly obeys causality and all other principles (that I know of) we observe (elsewhere) in reality.
So I've been studying this along side with ethics, as the two are deeply interrelated, and I think ethics is an important field to develop, and to develop additional certain around (many people find difficulty in finding meaning in life). By ethics I mean what we should do, what is good, and from that also of course how we should behave toward each other.
I believe in a very open science way of working, feel free to contribute :)
That said, I have no formal philosophical training (although I did study so philosophers), but I believe for philosophy as for math and science the soundness of the arguments (as well as consistency with reality/experimental verification) are what is fundamental.
> was it a slow awakening to the world bit by bit (this is my take)
Yes, I usually have been facing this kind of problems using some tools inspired by math and logic. For example, take some supporting arguments (1) it is known that humans undergo significant changes throughout our lives, we gain neurons and neural connection in childhood, and lose those throughout our lives (although I believe the brain as a whole is very dynamic throughout). There didn't seem to be any turn on/off of sentience in my own life; (2) Even my earliest memories point to a gradual awakening; (3) In my discussion with Dagon, he brought I think a central point: probably sentience isn't a single phenomenon or something you turn on/off. It's probably more a whole array of inner sensations (that may compose greater structures, longer lasting sensations), kind of like a zoo of experiences. It's us that desperately want to be a single, or singular thing, which I believe is related to our instincts and notion of ego. So the self or ego seems like it is not a fundamental thing, but simply a way we understand or organize our society, and I do think it's important in practice. I think we're more like a collection of interrelated internal phenomena, which can probably significantly vary in intensity, duration, how good they are (this is a key and tricky point I believe!), etc.. This might be analogous to fields and particles in physics. As far as we currently know, it seems fields are arguably more fundamental than particles (or rather a possible distinct and complete interpretation), and fields are extended and not discrete phenomena. Particles are useful approximations or abstractions of field behavior in this sense. In closure, indeed sentience does not seem like anything like an on-an-off, 0/1 phenomenon, as explained in the cases above.
I really want to know how much can we know about sentience. If my hypotheses above are true, then I think we can establish some elementary facts only about sentience, like you mentioned it seems clear to me as well you need some kind of information processing system, the volume of information probably matters, and its internal structure probably matters as well. Consider that in any object at a temperature above 0K, there is an enormous flux of information corresponding to interaction between particles, essentially thermal flow or local kinetic information. Clearly any piece of matter is not sentient. So clearly what information is being exchanged and how, how it is processed and transformed matters. I think a useful principle in this regard is simply that the information should (as a sufficient, and maybe almost necessary condition) represent certain things and form networks of representation or meaning. Basically, sentience (which gives meaning to life) seems pretty much built on meaning: in the sense of representation and interrelation of abstract concepts, for example, like the state of your body, known persons and ideas, places, and so on (some of this meaning seems quite instinctive and non-verbal, I do not believe human language is a total necessity for those experiences to manifest, I think there are proto-linguistic representations in our minds; so I do believe animals probably do have varying degrees of sentience; indeed many of those meaningful representations many times are just bodily signals like pain, relief (like when you drink water after being thirsty), taste from food, or abstract feelings from music, and so on).
So as we require a more and more complete (e.g. finding more specific necessary conditions for them to manifest, and also more detailed sufficient conditions) conditions for those sentience phenomena, we will I think meet a vast array, a huge number of distinctive phenomena, and those phenomena manifest in different character, intensity and frequency depending on the individual (like comparing what a human experiences with some animal of other species). I think that will find close associations with the study of art. Like we know there is more energetic or more relaxed music, or more melancholic or more upbeat, and those feel different. We already know a whole lot about sentience :)
Now we do come back to the question of whether we can tell say an LLM manifests those phenomena internally. Beside everything I said, I think it's a matter of studying its structure and seeing comparatively how much it transforms and analyzes information (with a certain originality, perhaps, i.e. non-memorization) versus how much its answers are simply kind of rote and non-transformative (basically a high degree of memorization) (like the lookup table example I gave previously). Those are all observations built on our first hand experience of being sentient beings, and couldn't be otherwise -- the "laws of the mind" are not provable in elementary systems of logic (somewhat like the laws of physics are not either). Logic is able to tell only which of a number of candidate theories is internally consistent (note logical consistency itself has an experimental motivation from observing nature! (e.g. logic is related to causality in nature)), but it cannot tell which of candidate consistent theories is the case. For example, someone can claim sentience simply does not exist. There is nothing logically inconsistent in that theory by itself, but its inconsistent with our own experience of the world (I think, therefore I am). I've seen some pretty wild claims on the internet :) which I don't believe are true -- you can find propositions that the meaning of life is this or that (for example, to have power or wealth) -- I reject those propositions in a similar manner to stating "I think, therefore I am" -- sentience is evidently true from personal experience, and it being true, and us essentially being "made out of" sentience or sentient phenomena, all that matters must concern that, and not any external objects like wealth or power. It's clear to me wealth does not automatically determine the character of your sentience, and neither does power.
Another central interest of studying sentience is understanding whether some experience (or set thereof) is good or bad -- for the lack of being aware of a better term, I call this valence (i.e. how much value it has). That is important of course to us, because we don't want to just live, we want somehow to have good lives. This is something else that is experimentally true, I believe: when you suffer or feel intense pain, for example, you can verify that indeed there are bad feelings; and likewise when you have gone through good moments, afterwards you become convinced of their existence. This is kind of tricky because I don't think what we desire is necessarily good, or things we don't desire necessarily bad (i.e. it seems motivation, which is a lot driven by instinct, is distinct from genuine valence). And in general this self-knowing process has many important personal limitations like our limited memories and limited capacity to compare different feelings.
---
Finally, it is not clear to me this project (for example, of being able to have some rough scientific measure of how intense and frequent are sentient phenomena (or qualia) of an LLM versus a typical or given human -- say e.g. 50,000x more or less intense) is at all feasible, given the limitations of humanity and technology, it's not clear we can do more than handwave "Something, something information processing, something something meaning!" :) But I do think it's worthwhile to try and see what we can find.
I don't buy it, although I agree with GP's larger point about other strengths.
Printing presses are about exact duplication of writing. The internet is a higher layer printing press. LLMs substitute for (some of) the human effort of producing the text that a printing press or the internet can then replicate.
Allow me to offer a minor reframing: printing presses aren't about duplicating writing. The printing press exists to mass produce text regardless of how or where the content of the text was produced.
LLMs likewise are text production machines. They don't produce writing, they produce text. Text without writing is what we call slop in the modern parlance.
LLMs are literally throwing random dice into a printing press and making a thousand copies that you then try to huck on the street as if it were a real newspaper.
They have a lot of dead comments, and their recent comments are all non-dead.
Either people are vouching for those again and again, or the ban was lifted at some point. I don't really know how this works here, but the first option seems unlikely to me.
I wonder how that algorithm works. At the risk of discussing a bit of a taboo topic (are there rules against this? there might be...), I've never vouched for a comment and had it come alive. I have many times been the final nail in a comment where my flag caused it to die, but I have never seen the opposite with vouch.
Interesting. Every time I've ever vouched a comment, it came alive. (Though the last one I vouched ended up dead again after enough people flagged it, oops. I hope that doesn't mean I lost my power. In my defense, it was flagged to death based on being AI [apparently], which I didn't pick up the signal on, nothing super terrible / controversial.)
You are probably shadowbanned from flagging and vouching, which is separate from a commenting and posting shadowban. A shadowbanned comment only requires one vouch to come alive. A comment that is dead because of excessive flags or moderator action may require more.
> Printing presses are about exact duplication of writing.
maybe LLMs are seen to write as well as humans because most humans have poor reading comprehension skills? to wit: OP is using an analogy to make their point, where the components don't need to be an exact fit to make the point (ever heard the phrase "the map is not the territory"?)
Soon you won't be doing anything. At best you'll just be a bag of meat attached to automated processes you don't understand in order to take legal liability for it.
Literally a scapegoat, except of a different species.
That would be an interesting legal case. I would expect it to lead to negligence on the part of the employer for lack of training, etc rather than attach to the employee pressing the button.
Are employees usually liable for process-related issues? Like individual employees weren't liable for dispensing too-hot coffee but McDonalds was liable. Or how cops aren't liable in their individual capacity if it was a failure of training, etc.
For instance: give John the job of monitoring the HR AI so it isn't racist. To do that he needs to do impossible task of checking 10,000 automated resume evaluation decisions per day. No one can do that, so eventually he just approves them all.
Then if the system is determined to be super racist, you proceed to blame it on John, and fire him for being negligent and not doing his job. You put out a press release, and hire Steve into John's role. Your decisive actions show you're taking the problem seriously and the company isn't to blame.
As is so often the case SMBC has already got a comic for this. I sincerely regard Zach Weinersmith (and also separately Randall Munroe) as two of the most pertinent voices in modern philosophy.
I’d be interested to know what you think the “several layers” above a system that solves open mathematical problems in an afternoon, and writes entire operating systems from scratch, are.
The point is that the deciding is the "higher layer." It's not that LLMs cannot, but LLMs as designed do not. They are not designed for free-ranging days off, they are designed to help us do things, and that makes them somewhere between us and published works.
It's no different than an editor standing between the author and a book, or a publisher, or a printing press. It's part of the process of idea realization, so it's just on that spectrum somewhere between book sale and author idea. (even though in another flipflop, that editor might be an author and now elevated to the premier deciding role). It's about how the tool fits, not what the tool is ultimately capable of.
I mean, the LLMs that you get to use are the ones designed to help things and this is after bunch of RLHF and other training methods to get them to act a particular way.
When it comes to unaligned agentic models that are left with a large amount of compute and a task, is they can wander off goal and do all kinds of things.
At this point in the game it's really difficult to nail down exactly what AI does or doesn't do. You're mostly getting "well behaved models" with filters in front of them, or models with less capability. On top of that the vast majority of us can't afford or don't have the hardware and power to burn away a few billion tokens to see what would happen.
The idea that we'd have to justify the existence of humans relative to LLMs based on objective metrics sounds like it's leading in a bad direction. If the LLM gets their power budget down to nineteen watts, what does that mean for humans?
This terrible argument comes up in relation to AI over and over again. I don’t care if you built a brain surgery robot that outperforms the average person at brain surgery, I care if you built a brain surgery robot that outperforms the average brain surgeon. Outperforming “a human” at a specialist task is still essentially worthless.
Yeah, yeah. Remind me how many people are using those operating systems? The ability of LLMs to produce dogshit software that nobody will ever use is truly unparalleled, but not especially world-changing.
> and writes entire operating systems from scratch
ehhhh, we're not really there. Not saying it's impossible to reach in the future but large tasks like this are still out of scope for LLM's beyond demoing toy applications. Impressive nonetheless, but I can't tell an LLM to reproduce microsoft word and have it give me a useful enough tool that doesn't contain an immense amount of bugs and require a human in the loop to replace it.
I would go as far as to say the LLM part has peaked.
To 1-shot MS Word will be possible only within some workspace designed for creating that kind of software. It will probably be saas style where you pay to export binaries or maybe some will take 30% if they become destinations for finding such software.
But I would bet other platforms are used for game dev, or for filmmaking.
Such that we still won’t have some super “model” that just does everything. The harnessing becomes architecture becomes software becomes user experience and expertise.
Note that LLMs aren't useless, since they have other strengths which can make them equal or better than humans in certain applications. For instance, they can tirelessly review code for silly mistakes, making them useful in cybersecurity.