03:43
I've been spending my time rather constructively as well. But I daren't say what I've been up to, lest everything I write gets cleansed again for "not being on topic"
So I'll give my contextually relevant two cents on AI. AI as it is right now is simultaneously overhyped and underhyped.
On one hand, it's the fancy new post-crypro buzzword all former NFT bros are using to describe what they're "pivoting" to, the thing greedy managers have wet dreams of that allows them to fire even more staff and have someone do free work for them.
Hell, given the absolute disgraceful state of modern scriptwriting, it seems AI has already taken over.
One of the big issues is that AI is being lobotomozed to hell and back, to the point where you can't even ask ChatGPT how to cook rice because it'll reprimand you for conducting a potentially dangerous activity
I'd much rather have an AI that unabashedly hurls insukts at me or teaches me how to cook meth, rather than an AI that tells me that the longesr five letter word is "twelfth" with 7 letters.
At the same time, AI is underestimated significantly. Just like an autist, it sucks at having good conversations, but is incredibly good at finding patterns in data. That's fundamentally what AIs are best at.
So a less-than-ethical corporation (i.e. all of them) could use an AI to predict which workers are most likely to unionize, to then strategically "choose not to extend their contracts"
Or governments could use AIs to correlate a person's writing style across different data sources and tell with pretty high certainty whether or not two different texts were produced by the same person through word choice, common mistakes and other mannerisms.
Or the fact that generative AI made "this image/video of me isn't a real photograph" a legitimate defense in court
Who cares if there is a video of you putting a hat on a goat or similar extremely unethical acts? "It didn't happen. Don't trust your eyes. It's all just AI"
AI safety is as much about making sure the AI doesn't fuck up and decide to hurl your self-driving car off a cliff because "lmao I discovered a shortcut", as it is about making sure humans don't use AI to be even worse humans
Because if there's one thing that is absolutely constant about new technologies, it is that people will use it for porn, scamming others and to oppress others. In that order. Two of these already happened and the third is on the way.
AI is also a technology that explicitly only aids the already wealthy. AI models require immense amounts of storage, far more than any single person could ever be able to afford, or even just have the infrastructure to make usable. Training an AI model has become so complex that companies like Google or Amazon are making their own TPUs to speed up the process.
In other words, no matter how "powerful" AI models might be in the future - the models owned by "the elite" will be several orders of magnitude more powerful than anything the populace will be able to own.
The average consumer has access to hard drives of roughly 8 TB per disk. Even a NAS with 8 disk slots "only" gets you to 64 TB per NAS, without any redundancy of course. Even if we presume you can store ~1 PB, Google stores ~ 2.500 times as much.
2 hours later…
« first day (4741 days earlier) ← previous day next day → last day (371 days later) »