> But if you could send an LLM back to 1950 and train it on the pre-existing literature, you couldn't have it generate The Shining, A Scanner Darkly, Piranesi, Perdido Street Station, The April Witch, or Starship Troopers.
Why? I don't see a clear reason why it couldn't that does not invoke the mystical "human spirit".
To be clear, I don't expect a single LLM to create such a book, at least not in anything resembling current models, but I can absolutely imagine a long multi-agent setup that would. Given enough money, I'd set up a mechanism whereby author AIs generate books, separate agents act as editors and critics to iterate on a small fraction of the best of these drafts to polish them, and then publisher AIs choose an even more minute fraction of these to publish into "the zeitgeist", which will then have other agents offer further commentary on these books, which would then feed into the context (or even fine-tuning/training) of the next generation of LLMs. I expect that a sequence of e.g. ten generations of swarms of AIs, each perhaps of ten thousand agents, could arrive at a ground-breaking book with actual literary merit.
In my company I have a very good conversational dynamic with my team. So in my first forays into LLM usage (the past couple years), I began trying to designate models as team members and mimic similar conversations and decision making processes that we have at work, just as an experiment. Somebody to brainstorm together who likes going on tangents, somebody who unapologetically removes unnecessary cruft and cuts scope, a few other people with specific tendencies and personalities.
The results were completely awful. LLMs couldn't even remotely mimic these personalities. The conversation was all trying to please no matter how contentious I tried to instruct them to be. LLMs just seem to struggle a lot with arguing and being contentious. Which probably makes sense based on training.
I feel like these types of factories in practice don't work out too well.
First off, I just one to say that while I'm strongly on the other side of this argument, and in favor of heavy AI use for dev work, I really appreciated the article and the way it makes its case.
My main point of contention is really with this:
> House House is not a studio that couldn't afford to hire a specialist for a few days to look into this problem.
I don't understand why there's this expectation that we have to throw people at every problem. Let's say that a dev faced with a difficult technical challenge in an area that they're not familiar with had instead gone to a library, and came out of it after 5h with a solution. Or alternatively, that they googled it and by having rolled the proverbial critical hit found an obscure blog post from 2005 with the exact solution to their problem. Would there really be a big difference? Why should we pay each other to do stuff for us that we can easily do with tools at a fraction of the time and cost?
Can anyone make a strong argument against using AI for dev work that would not immediately lead to needing to bring back all the elevator operators?
To me the question just doesn't make sense, being a category error, like looking at football players and trying to identify the position of the team spirit, or looking at a country from space and asking "Do I need to zoom in further to see the GDP?"
I can't conceive of sentience as anything but an emergent property of a "running" system that can only be observed by seeing it in action, rather than by looking at its static components.
But even if your bar for sentience is completely simulating a brain and you agree that it is possible in a few decades, isn't that sufficient for some rational civilization-wide panic? As a reference, in the 3 Body Problem books, people know that superintellegent aliens will arrive in about 400 years and are trying to vigorously prepare for that; shouldn't we do more to prepare for synthetic sentience even if it'll take decades?
I agree with everything else you said, but that part rang weirdly to me. I don't see anything AI-specific here. As I see it, other than maybe some core gnu utils (and I'm not even sure about those), software is meant to continuously evolve and improve. And especially for passion projects, why should I want to feel "quite finished" about a project?
Agreed, but I just wanted to say that essentially every single thing anyone owns is, and has been from the invention of tally marks on clay tablets, an entry in someone's ledger.
This is too reductive. There's been real development behind what it means to own something. What's written in a ledger is just an inscription it has no meaning outside of its social context.
Our social understanding of property has changed. John Locke's ideas on private property are so fundamental to how we think about ownership that we don't even realize there would have been another way of thinking about how we own things.
Of all the thousands of things I own, as far as I know, only my car and my house exist in anyone else's ledger as belonging to me. Am I missing something?
I would find that hard to believe. You have a bank account, surely. A life insurance policy? Airline points? An account with the government social security office for your pension? There are probably dozens (if not hundreds) of things of value (even if the value is near zero) that exist in someone else's ledger.
But equally hundreds if not thousands of things a typical person owns isn't recorded in any ledger. The personal laptop currently in my lap, my clothes, my furniture, my pen and notebook, the cash in my wallet
Saying all ownership are just ledger entries is just as reductive as saying no ownership can be reduced to ledger entries
Presumably this is like many other graphics tool vulnerabilities this works by the attacker persuading a user to open a file that exercises the bug. Not terribly unusual.
I'm no expert but so much of Windows is implemented as services. svchost.exe is probably involved somehow. Like it wouldn't surprise me if Copilot is implemented as a service and it talks to Paint via IPC messages or something
It would make sense that it could listen for messages like "your generated image is ready!" and pass bitmap data to Paint. And media like images are a common source of vulnerabilities since you're basically passing binary data into the software.
This site absolutely does jokes, it just doesn't beat horses into bloody pulps with the jokes the way reddit does. We have a higher standard. A joke has to actually be funny, and not some tired reference.
Not to imply any particular comparison to less serious websites where the karma system over time may have resulted in a disproportionate surplus of one liner jokes and other spammy low quality posts.
Why is that a mixed metaphor? I'm very used to large initiatives/programs with several large milestones, each involving multiple projects. I don't see any issue with it.
The milestone they are reaching is the formalization of The Stacks Project. It began whenever they started prompting an AI to do that or started preplanning.
The metaphor of milestone is a goal reached within a larger process. I think that, yeah many people will not find a problem instances of this sort of (mis imo) usage but AIs to create a sequence of these things and that tends to create a tone to AI writing that many people can feel even if they can't point to particular problem usage.
Why? I don't see a clear reason why it couldn't that does not invoke the mystical "human spirit".
To be clear, I don't expect a single LLM to create such a book, at least not in anything resembling current models, but I can absolutely imagine a long multi-agent setup that would. Given enough money, I'd set up a mechanism whereby author AIs generate books, separate agents act as editors and critics to iterate on a small fraction of the best of these drafts to polish them, and then publisher AIs choose an even more minute fraction of these to publish into "the zeitgeist", which will then have other agents offer further commentary on these books, which would then feed into the context (or even fine-tuning/training) of the next generation of LLMs. I expect that a sequence of e.g. ten generations of swarms of AIs, each perhaps of ten thousand agents, could arrive at a ground-breaking book with actual literary merit.
Give me the budget to prove me wrong :)
reply