We’re a software development firm based in Pakistan and we created a brain that is only 400KBs; Libbrain4 is a c/CUDA-based library and it focuses on learning like a human does wherein the brain actively breaks down projects into smaller functional portions that it then solves akin to a human. The program follows a simple self-improvement, reinforcement rule that does edits and tests on every iteration. We started on the third of November this year and its able to get an independent raisings score of .4 as well as get a .07 on PPO. Its score actively improves on every iteration at a rate of .01.
Goals
We want to be able to do public benchmarks on 10 seeds on Atari 100k against DreamerV3,, BBF, and EfficientZero and measure the speed at which it learns relative to human beings as well as to generally improve the learner.
Proof
We’ll use human data and do comparisons to the Atari 100k table with all of this being available publicly on our website.
Funding Usage
The current model runs on my own laptop, an RTX 3080. Our minimum target is $1,500, which we’ll purchase two GPUs and if we get $12,000, we’ll purchase 64 vCPUs & L4-class GPUs. Currently, all of the funding for this project is out-of-pocket.
Team
I’m Rana Aitsam. I’m the founder and CTO of KHive and have been in the Web3 industry for a while. Our company has built a number of software and hardware projects for our clients ranging from AI agents to complex recognition software. We’ve been working in the industry since 2017 and Libbrain4 is our effort towards creating artificial life.
There are no bids on this project.
Comments
Omair Khan
about 23 hours agoI'm intrigued at the concept
Rana Aitsam
about 10 hours ago@omairkhan2194 thanks we are now testing with a fighting game as well, we will publish those results soon
Hasnad Munir (Clara)
about 24 hours agoReally interesting project! Hope it gets funded
Rana Aitsam
about 10 hours ago@hasnad10 every $ counts, please tell your AI enthusiastic friends about it too