22 ms·

And where will these models run? And who exactly will create the content? Your personal AI will just consume a shitload of content that the powers with most com

by mola 3y ago

And where will these models run? And who exactly will create the content? Your personal AI will just consume a shitload of content that the powers with most compute generates. Reality is gone, it won't even be accessible via the shadows on the cave wall.

r3trohack3r 3y ago
> And where will these models run? Locally, on my devices. Where property law says I own the fine-tuned model and the device. > And who exactly will create the content? Anyone and everyone? I don't understand what you mean. What you just posted here is content. You produced content. I assume you will continue doing that moving into the future. > Your personal AI will just consume a shitload of content that the powers with most compute generates I think that assumes a random walk through the internet? Content discovery probably shouldn't be random? As your AI curates content, you give it ongoing feedback about the quality of the content it gives you, (feedback based on your highly biased definition of "good" and "bad" content). Your AI learns what is "good" to show you and what is "bad" to show you. As your AI finds "identities" and "content channels" that produce reliably high-quality content that your highly-biased and highly-aligned AI can curate for you, I'd imagine your AI would prioritize searching those content spaces for quality content first. For example, I'd like to think my personalized AI would learn to surface up interesting conversations and content from Hacker News. As content channels get "overrun" by noise, the AI is filtering out the noise, the signal-to-noise ratio drops, and the AI slowly lowers the priority of those content channels in favor of channels that consistently produce content you find valuable to consume. The key here is that the AI is ultimately accountable to me, overseen by me, controlled by me, fine-tuned on signals generated by me, etc. Imagine you had an army of 1 million personal executive assistants who could scour the world looking for information relevant to you, and give you a curated feed of useful information on an ongoing basis. That sounds like a pretty cool future to me. And if someone wants to spend their time squandering their wealth to accumulate enough hardware to run 1 trillion personal executive assistants to generate a bunch of valueless (or value-negative), coercive, uninteresting dribble on the internet? Cool. Let them squander their wealth. My executive assistants will filter that out for me.

by mola 3y ago

r3trohack3r 3y ago

> And where will these models run? Locally, on my devices. Where property law says I own the fine-tuned model and the device. > And who exactly will create the content? Anyone and everyone? I don't understand what you mean. What you just posted here is content. You produced content. I assume you will continue doing that moving into the future. > Your personal AI will just consume a shitload of content that the powers with most compute generates I think that assumes a random walk through the internet? Content discovery probably shouldn't be random? As your AI curates content, you give it ongoing feedback about the quality of the content it gives you, (feedback based on your highly biased definition of "good" and "bad" content). Your AI learns what is "good" to show you and what is "bad" to show you. As your AI finds "identities" and "content channels" that produce reliably high-quality content that your highly-biased and highly-aligned AI can curate for you, I'd imagine your AI would prioritize searching those content spaces for quality content first. For example, I'd like to think my personalized AI would learn to surface up interesting conversations and content from Hacker News. As content channels get "overrun" by noise, the AI is filtering out the noise, the signal-to-noise ratio drops, and the AI slowly lowers the priority of those content channels in favor of channels that consistently produce content you find valuable to consume. The key here is that the AI is ultimately accountable to me, overseen by me, controlled by me, fine-tuned on signals generated by me, etc. Imagine you had an army of 1 million personal executive assistants who could scour the world looking for information relevant to you, and give you a curated feed of useful information on an ongoing basis. That sounds like a pretty cool future to me. And if someone wants to spend their time squandering their wealth to accumulate enough hardware to run 1 trillion personal executive assistants to generate a bunch of valueless (or value-negative), coercive, uninteresting dribble on the internet? Cool. Let them squander their wealth. My executive assistants will filter that out for me.