I took a break for 4 years, but I'm researching where I left off at in my code I was working on.
I'm solving many problems fast and have a clear and full roadmap. I'm going to attempt to see where it leads to still. The idea is based on stemming from exact matches in a network as deep as possible (I'm not going the backprop route like modern AIs). I know some of this has definitely been tried out there, but there's a lot of things to it and I don't think anyone has successfully implemented it all. I really haven't seen it's limits and I see how it can all be implemented. I have a complete plan already to make everything work while staying efficient and on GPU. It's definitely a little silly to work on it when we have advanced AIs already and a clear roadmap, but the architecture I'm working on could be better because it's fully transparent and extremely tiny code, and can train / upgrade its intelligence even after training, and fast training. Most "near-SOTA" AI code I see online and in the Large Text Compression Benchmark page are very large. A lot of this, but not all of this, is due to documentation and ex. TensorFlow usage, but that's also not good at all. Even nanogpt and tinygpt github repos are actually like 20 files each 200 lines long under the source code folder. Very weird. I'm aiming for a very tiny code of say 200 lines. I'll see how far I progress. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T6cf3be509c7cd2f2-Mfd0067b19715d9e0e8445a8f Delivery options: https://agi.topicbox.com/groups/agi/subscription
