Who here has been working very deeply on this actually trying to test it or pin point it? I mean how do you make the AGI talk on its own and digest what it gurgitates in its thought train? Data compressors do it, but what are you doing now Matt? What's the plan / current status? We all get a lot, AGI will be given primed nodes and Generates desired/related output data and equalizes closer to optima haven, but do you have any further structure to add here or neural architecture scheme?
I'm looking for more info here, others's structures, real testing, and what 'ties it together'. We seen all these data compressors and GPT-2, and they all can generate related data (desired output for users aka generate the future). Anyone know how to turn these predictors into something more AGI? My basic plan is to make a really, really, good predictor for Text, using Lossless Compression for evaluation obviously, and it should 'talk' like GPT-2 and even better. Once I achieve this, I will manually hard-tune nodes in its brain like words like 'love' 'humans' 'food' and these nodes will leak dye to other nodes, as does energy leak as well. I'll also have it prune unused memory, and only store interested-in readings (using RL nodes I primed and their related discovered likings (it may decide to like dogs once it realizes they are similar to 'humans' - the node I hardwired). It'll read the Internet sources it likes most, using 'actions' to surf the web. It'll be trying to come to equilibrium in its brain of text, talking about recently heard and favorite nodes, hearing itself, then storing what it says in its head. It'll also ask us questions and be asked questions and have attentional weights to multiple inputs, as do words in the last 20 words heard, and as does multiple sensory votes act as multi model judges too. And I believe I have a really good neural architecture to recognize/activate the related nodes and understand the input (last 40 words read). Although it may reply to the last words heard (like asking a scientist 'what do you think of golfing?') making you 'automatically' waste time answering them, it usually stays on its own hardwired nodes and has a Dialog context Always Activates / primed. So all day long I talk about my desires/questions/context. 'AGI' is just trying to predict/output the right text to get eternal life. Of course this is the first stage, the 'body'/'vision' will be very useful, but we shouldn't need it right away, all life problems need R&D but you can find many many many solutions the more data you have, as you already have a lot of data and hence know what will happen without testing anything. That's a big chunk of my plan, with actual criteria and real algorithms already in existence. What's your plan? t better be as big as mine above or I'll be unimpressed :D. So what your rundown, ? ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tb135fde0a846b51f-M20c62401a32fbe0dbce58049 Delivery options: https://agi.topicbox.com/groups/agi/subscription
