Once AGI is made, it will be trained on all our data in a few weeks by the time we reach 2029. Right now it takes some weeks to train GPT-3 on 570GBs of text, give or take a month or two. It won't take long. Yes we'll have a ton more data by 2029 at our projection but it's good... Then this smarty pants AGI will be cloned 1,000,000 times, nearly instantly (and we have enough computers to run 1M GPT-3s!) and they will all work together nicely on slightly different tasks (ASI related), cloning already happens as we speak with AI parameters. 1M is how many humans probably work on AGI. This won't just double our ASI progress, it will feel like 3M workers! Because AGIs don't need to sleep or eat or exercise, think of bad memories (can erase them), worry about death much, etc. Our 3M worker progress movement will make now 3 years of progress in 1 year on ASI due to having 2x more workers! We will optimize the AGI code to run 4 times faster, making their 2x speed now 8x as much work a year. That's 9 years per year, we will reach ASI level in just a year or 2! The AGIs will run fast in their brains unlike the human fleet, meaning they will be able to test changes their code regardless of human slowness or requiring a body to do so because it is just testing code. The ASIs will also improve their intelligence to recognize things better, ex. walk = w a l k, WALK, KLAW, run, wajjjlk, to go very fast, 'vision of such', etc. More speed, large data, and higher intelligence are HUGE factors that affect how quick you can solve problems. SO: the code speedups and intelligence will move them so fast that hardware probably won't keep up. They will probably design their own way and will tell us how to make nanobots to attain large amounts of resources/ data/ sensors/ manipulators/ brain/ energy.
BTW I think I have learnt that the whole algorithm is matched based, so when you see part of the NEURON ex. "walkin?" or "walked very ?" or "relatedwordmathced(ex. walked very) ?" or "dog cat ape fish rat ?" (priming, either same or related), it finishes it and doesn't say it again either. It is just matching it to a known memory and then saying it again in the future or the rest of it that remains to be said! SO: We will have enough computers to run 1M AGIs, most the work will be them optimizing their code at high speeds and their intelligence too. It's not necessarily our job to make more better computers. The people will do it. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8b81e7588ed2cbad-M887f44231f33a3202a34708c Delivery options: https://agi.topicbox.com/groups/agi/subscription
