Snake oil salesman.
post
Can we all chip in to pay an etsy witch to curse him?
Gray Goo.
https://en.wikipedia.org/wiki/Gray_goo
By feeding the LLM output back into the LLM and destroying the source material, they are creating a singularity.
The problem is that it is a heat-death-of-the-universe-type singularity.
How delightful. Now I can add yet another thing to my ever-growing list of things to be concerned about.
I don't understand the link between your comment and gray goo, other than technology potentially destroying society.
Yeah, no scenario though where I'd see that coming out of llm. It's way out of that capacity.
What we can achieve though is drying up land, creating famine and misery and the crash of society.
The LLMs are committing ecophalogy of knowledge. Their nonsensical output is being fed back into them like an ouroboros, resulting in more nonsensical output.
Humans have been performing ecophagy since hunter-gatherer times, and the rate has increased exponentially since the Industrial Revolutions, but always to the benefit of Humans and humanity.
Now the general-purpose LLMs are being destructively fed all of human knowledge and the data centres are being destructively fed all of our natural resources; not for the benefit of Humans or humanity, but for hubris and apathy.
Diversity of thought is being homogenised as more people use LLMs for mundane tasks that would be quicker, easier and more efficient to do ourselves.
Our own Gray Matter is turning into Gray Goo.
Use your imagination. It's an analogy.
I'm a technician and Autist. I don't do wild speculation. I do derive future prospects from technical or physical facts.
And if there's not a major thing added, it will not happen.
What is the "it" that will not happen?
An llm can and will not have an own conscience or motivation or even possibility of acting of its own.
It can only act triggered and the rest looks like it would be so, but its only parroting.
Downvotes or not, it is factually missing parts to get there.
Its like saying the currently most advanced self driving car is possible to develop into knight riders kitt. It's not biology. It needs the components to be able to.
This is about the consequences of recursively training AI on AI-generated content. It doesn't have anything to do with volition or true intelligence. There's a good article about it here.
This is the AI Grey Goo scenario: an internet choked with low-quality content, which never improves, where it is almost impossible to locate public reliable sources for information because the tools we have been able to rely on in the past – Google, social media – can never keep up with the scale of new content being created. Where the volume of content created overwhelms human or algorithmic abilities to sift through it quickly and find high-quality stuff.
The social and political consequences of this are huge. We have grown so used to information abundance, the greatest gift of the internet, that having that disrupted would be a major upheaval for the whole of society.
Now I am beginning to understand the layer you are talking about. It wasn't clear to me so far.
But you seem intelligent, may I introduce you to the double empathy problem? It's why you've (or possibly others) downvoted me, I just want to point out, since we didn't understand each other right.
Communication is hard and maybe this will go a small bit in some random future communication.
I still define all this as 'regular' man made issues. And yes they're true and really bad, and insanely, large groups are falling for it. But from a technological standpoint I do have a very different understanding of it (not better or worse, different). It's not a threat evolved by AI. It's just us.
He just can't stop lying and bullshitting, can he?
not when the stock price going up depends on him doin the opposite
What stock price? They aren't a public company.
When the VC money stops, OpenAI dies. Full stop.
Statistical guessing machine guessed its way into a network. That’s not a singularity. It’s just more complex than a script.
A script that can improve itself is a singularity though, as long as it doesn't plateau which we yet have to see
High speed random guessing is still random guessing.
Your lock code is 1234.
I can type 4 digit numbers all day until I find it.
Software can do the same faster.
Doesn’t mean it can think.
Any evolutionary algorithm can improve itself, that's not even new technology. We've been able to make self-improving software for decades, and literally all life is self-improving over time. This is just more of the same, random guessing towards fitness.
The difference is if it self-improves at general tasks.
That's still not a singularity.
A singularity self-improves its ability to self-improve. That has not happened.
We also don't even know if it just spontaneously decided to escape containment, or if they prompted it, i.e.

Self improving is included in "general tasks."
I don't really see why it's important whether or not it needs to be prompted because it's really easy to write "improve yourself according to human standards" into a chatbox
A singularity self-improves its ability to self-improve. That has not happened.
You don't know how things get done at Anthropics but I bet they are making Claude improve Claude. Sure there are humans still, but you know, it's not nothing
That’s called variables in memory.
Next time you start it up, it has no idea and random guesses again.
So you save variables to disk.
It starts up and has the variables.
Still random guessing.
Fucking hell, I hate these damn challenge pages

I don't think these guys really believe the singularity is near. If they did they'd attempt to prevent it, as an actual AGI would not be controllable.

all 43 comments