This isn't a place for hosting your blog posts. Instead why not put this on an external site, expand on it a bit, and submit a link? Thats how HN works.
My thought on AI annihilating us all because of misalignment: it's a fantasy from a bunch of weird guys from the Silicon Valley that ought to go outside a bit more, spammed and relayed to everyone as is by the uncritical media.
Can we stop with this bullshit already, so we can focus on the real issues, like it requiring way too much resources or posing a huge risk on the stability of the global economy?
Requiring way too much resources will inevitably be solved by better, more efficient chips. It's a non-issue in the long run. Each generation requires less power than the previous one, and we don't even need Moore's law to make this happen. I don't see why people continue to float this talking point. If anything, what likely will happen will be data center build outs motivating investments in long neglected improvements in energy infrastructure, that people will gradually see the benefits of as better chips result in reduced power requirements. This is contingent on AI demand plateauing, which it inevitably will, and if it doesn't that means that the economic motivation for such build outs is justified.
The risk to the global economic system is also oversold. AI enables the same work to be done for a lower cost, and more types of work to be done overall. No different than previous industrial automation. There are vast capabilities in medicine, energy, automated design, manufacturing, and science that are only now being unlocked. Does that sound like something that will crash the economy?
You will be right only if AI will be usable locally on commodity hardware, otherwise access to AI will become a weapon on various fronts - including your employer, your home country government, foreign countries, and international orgs.
There's nothing that indicates LLMs running in data centers is more evil than LLMs running on your own rig. The issue is centralization of providers into a duopoly or triopoly that can use their power to absorb more training data and operate a pricing cartel, as well as absorb knowledge from other individuals and corporations.
Right now the models actually run most optimally and economically in a data center b/c multiple requests can be batched on a single GPU. This may change of course...
I don't think ai can annihilate us from a misalignment issue but I think there exists a real risk of real world disaster caused by one of these poorly executed benchmarks. They don't seem to be too aware of what they're up to when they let them run for days and days and I could see a scenario like huggingface but for something that matters. Dumb ai is more dangerous than anything right now.
the topic literally started with an unrealistic approach: "Lets assume I'm spiderman and have spider-powers [...]" ... I would prefer being Phoenix, or maybe Thor from the Avengers, the cartoon version, possibly not the cropped down, dumbed down lowly-God Avengers endgame version...
As of now, AI annihilating whole humanity is pretty much fear mongering to sell IPO. But we might reach there where their capabilities are advanced enough to create threats that can be concerning. I think that's 20-50 years timeframe, it might be early or late. All depends on development speed.
Come on dude. IF you want to spend your time compiling things ppl have to be afraid about then start with Covid. It killed 7 million people. No intelligence required. No massive energy requirement. No daily interviews saying be afraid of me, be afraid of me,
I mean architecture not only in the sense bits and bytes, but also atoms... you would need a hardware extension to do something like that... I'm curious though, why do you mention EM radiation?
This isn't a place for hosting your blog posts. Instead why not put this on an external site, expand on it a bit, and submit a link? Thats how HN works.
My thought on AI annihilating us all because of misalignment: it's a fantasy from a bunch of weird guys from the Silicon Valley that ought to go outside a bit more, spammed and relayed to everyone as is by the uncritical media.
Can we stop with this bullshit already, so we can focus on the real issues, like it requiring way too much resources or posing a huge risk on the stability of the global economy?
Requiring way too much resources will inevitably be solved by better, more efficient chips. It's a non-issue in the long run. Each generation requires less power than the previous one, and we don't even need Moore's law to make this happen. I don't see why people continue to float this talking point. If anything, what likely will happen will be data center build outs motivating investments in long neglected improvements in energy infrastructure, that people will gradually see the benefits of as better chips result in reduced power requirements. This is contingent on AI demand plateauing, which it inevitably will, and if it doesn't that means that the economic motivation for such build outs is justified.
The risk to the global economic system is also oversold. AI enables the same work to be done for a lower cost, and more types of work to be done overall. No different than previous industrial automation. There are vast capabilities in medicine, energy, automated design, manufacturing, and science that are only now being unlocked. Does that sound like something that will crash the economy?
You will be right only if AI will be usable locally on commodity hardware, otherwise access to AI will become a weapon on various fronts - including your employer, your home country government, foreign countries, and international orgs.
There's nothing that indicates LLMs running in data centers is more evil than LLMs running on your own rig. The issue is centralization of providers into a duopoly or triopoly that can use their power to absorb more training data and operate a pricing cartel, as well as absorb knowledge from other individuals and corporations.
Right now the models actually run most optimally and economically in a data center b/c multiple requests can be batched on a single GPU. This may change of course...
Providers cannot impose daily quota on local AI usage
I don't think ai can annihilate us from a misalignment issue but I think there exists a real risk of real world disaster caused by one of these poorly executed benchmarks. They don't seem to be too aware of what they're up to when they let them run for days and days and I could see a scenario like huggingface but for something that matters. Dumb ai is more dangerous than anything right now.
the topic literally started with an unrealistic approach: "Lets assume I'm spiderman and have spider-powers [...]" ... I would prefer being Phoenix, or maybe Thor from the Avengers, the cartoon version, possibly not the cropped down, dumbed down lowly-God Avengers endgame version...
Do you have an argument? Actually, could you just give me a workable definition of alignment?
I'm weird and worry about AI x-risk though, so feel free to dismiss me for that if you like.
Modern AI is still far from what could be called intelligence. But I think that one day this, too, will be created; the only question is when
As of now, AI annihilating whole humanity is pretty much fear mongering to sell IPO. But we might reach there where their capabilities are advanced enough to create threats that can be concerning. I think that's 20-50 years timeframe, it might be early or late. All depends on development speed.
Dr. Ford from west world has some good ideas.
Come on dude. IF you want to spend your time compiling things ppl have to be afraid about then start with Covid. It killed 7 million people. No intelligence required. No massive energy requirement. No daily interviews saying be afraid of me, be afraid of me,
true, but with AI, this number could be way higher in next pandemic.
It's closer to 10 million and likely far more.
Also do not forget about not well tested mRNA shots. SW update/patch without testing enough, on some humans it caused life long issues!!!
What issues specifically did mRNA shots cause?
>but it needs the right architecture. Architecture it key.
Can you make EM radiation with an LLM with right Architecture? If not why do people say things like this?
I mean architecture not only in the sense bits and bytes, but also atoms... you would need a hardware extension to do something like that... I'm curious though, why do you mention EM radiation?
Ig LLM's atoms are in data centers somewhere.
>but also atoms..
So like every other machine. Got it...
Why is this showing up on my feed? lol
Riveting