Let me do a thought experiment, say a human intakes text per minute, how much do they read in a lifetime, let's see:
74 words per 10 seconds is what I think is reasonable for a human to read, based on my self test just now. (Could be lower or higher in reality, even for me). So given that we sleep and stuff for 12 hours of the day, 319,680 words per day if constantly reading or thinking to self. So let's cut that in one quarter or so since ewe don't always read when awake or think or think of new data. 79,920 words per day then. This is in GBs, about 479,520 bytes, or 479KBs (lol). For 50 years, that is 8,751,240,000 bytes, or 8.8GBs. So Matt's number of 1GB is about right if we want say a 25 year old, where 4GBs may be still 1 quarter like 1GB if we assume humans actually think of the same text in the day and don't read so much or see so much new data. This is only really for text though, so it's still an estimate overall. But GPT-4 is only text and we can see how human it feels on just that even without being fully AGI. So yes, given just 4GBs of text, one should be able to make an AI sound human. GPT-3 used 40GBs of text, I think this comes close actually. It sounds human likely on 4GBs or 1GB, just needs a better algorithm and way to train it, etc. So in that regard we are close-ish to being there, it does at least feel human on 4GBs of text just needs some more work. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T592788a3fedc3e71-Me0b38665bf0ae6e11ed67b81 Delivery options: https://agi.topicbox.com/groups/agi/subscription
