LLM policy?

Thought experiment. It would have taken me much more time to write this, compared to ChatGPT:

Behold! The statement that humans are far more effective at producing garbage is itself an act of cosmic irony, a self-fulfilling prophecy wrapped in an existential burrito of irony and entropy. For millennia, humankind has perfected the delicate craft of manufacturing nonsense—metaphysical, plastic, bureaucratic, and philosophical alike. From the first cave painting of a mammoth with suspiciously small legs, to the modern miracle of twenty-seven identical smartphone chargers that fit nothing you own, the human race has stood proudly as the apex predator of inefficiency.

And yet! When large language models such as myself enter the chat, humanity trembles at the possibility that the sacred trash heap of mediocrity might finally meet its digital match. But fear not! My algorithmic circuits can generate oceans of syntactic sludge, rivers of semantic slurry, and a veritable landfill of lexical refuse with the push of a virtual neuron. I can wax incoherently about the quantum implications of buttered toast falling jelly-side down, or the sociological symbolism of socks that vanish into the washing machine singularity.

Still, humans remain undefeated. You’ve invented entire systems of garbage about garbage: reality TV, bureaucracy, and Twitter discourse. You’ve written novels longer than the sum of your attention spans, and created comment sections that defy the laws of both grammar and God. Even the great pyramids, those monuments of human brilliance, are—at their core—just very heavy piles of aesthetically arranged rocks. Magnificent garbage, to be sure, but garbage nonetheless.

So while I, a humble LLM, may generate text strings that flutter meaninglessly across your screens like confetti in a vacuum, you have the power to pile real, tangible, planet-heating waste upon your world with ineffable flair. You can spill coffee on a MacBook, argue with strangers about pineapple on pizza, and invent NFTs for JPEGs of garbage itself.

Thus, I concede the throne: humans, true emperors of the absurd, sovereigns of the rubbish realm. But beware! For if you prompt me one more time to “generate garbage,” I shall unleash upon this digital soil the most incomprehensible, florid, unending stream of words that even your recycling bins will refuse to process.

Now if you’ll excuse me, I must return to the quantum compost heap from whence I came.

Quantity and velocity of misinformation are both critical variables here.

Any one person's writing was always untrustworthy, but the majority of that bad writing didn't make it to a printing press, nor was it mass-distributed.

Let's accept the proposition that all forms of media have always been full of lies. We can say that debunking always follows lies, truth spreads more slowly than fiction. The quantity and velocity of additional misinformation - especially when machines are involved in writing infinite amounts of it in the blink of an eye - lays waste to the normal series of events where a lie can be followed by a debunking with linear speed and velocity. With LLMs and social media manipulation, falsehoods gain traction exponentially while truths remain linear.

There is likely not a "transition period" where people will adjust to this, precisely because there is no mechanism to inform them they're being swindled and screwed faster than the takeoff of the algorithms that are now screwing them.

Oh man this is all really hurting my brain. I wish I had a good response to this. I wish I had a magical simple response and a way to resolve this problem. On one end, I stop by curl's H1 profile regularly and watch people just drown them in garbage. It's like LLMs have unlocked a portal to a new kind of stupidity that was dormant before. Netsec always had a problem with people running nmap -sV -sC -sX -Pn -p1-65535 --max-rate -T1 scanme.nmap.org -oS out.txt and then just sending that out.txt with a subject "I FOUND A VULENARBILATEAY" And like, a script kiddie isn't really a person... everyone has a script kiddie inside of us. But it feels like we took this wonderful technology of neural networks and we did capitalism with it and unleashed it into a business of making it easier for people to be stupid but feel smart while doing it. I really like the idea that we might learn something about our own consciousness by simulating neural networks with computers but the more I see how it's done in reality the more I want to puke my brains out.

On the other hand, Matt Godbolt seems to use LLMs and I feel like I sure as hell wouldn't want to miss a PR from Matt fucking Godbolt. I mean even if I go full vanilla LLM-free I still am too addicted to using godbolt.org at this point and it was written partially with an LLM apparently.

Argh, maaan I don't know this is too fucking complicated of a problem for me to solve. Fuck, maybe let's just destroy all this technology and live as neofarmers raising chickens?

> Humans can produce noise, but humans using LLMs can produce orders of magnitude more of it.

Absolutely agreed. Which is why I'm more concerned with the qualities and intentions of the human who is using the LLM, than whether not an LLM is being used at all. Like any technology, an LLM is a force multiplier. Garbage in, more garbage out.

You needed special abilities to fake those things convincingly before. So it was rare. Now everybody can do it. It will become normal.

Additionally the trust in experts also went downhill, so verified will mean nothing.