It is interesting, AGI needs some things if you don't do a complete brute force to get an AGI program. So depending on the implementation, ex. GANs, Transformers, etc, you will be using some tricks, more or less. But there is a sweet spot of tricks we want to use, like: more data, RL, casualty, related, priming, gaps, delays, exponential, multi-modal, categories, to name many of them. So I think finding the right implementation is sort of easy, I think the issue is more about the way you do it....ex. using backprop is one way to update weights, I just do it another way.
BTW I'm looking into GPT more, I'm combing some my files and asking others, so I can do both paths at once. I really like GPT. I know looking into GANs may be an interesting approach. I haven't seen as good results from GANs, nor do I find how they work exactly interesting, it seems they can 'use' Transformers. The link below shows what they can do, and many these things DALL-E can do so idk. As far as a human like base to work on, I am saying GPT would be an ideal start to suggest how to get closer to AGI, not that you 'can't' work on possibly better but currently less better approaches. https://analyticsindiamag.com/gans-biggest-breakthrough-in-ai/ ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5b614d3e3bb8e0da-M2e41e39bad96b37cc5f6e9bf Delivery options: https://agi.topicbox.com/groups/agi/subscription
