You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$89,000 left to reach the funding goal.
I'm developing ACS(Adaptive Computational Substrate) which is an experimental AI architecture completely different from LLMs. Im currently investigating whether intelligence can be made more computationally efficient through adaptive allocation and reuse of computation
The primary goal is to determine whether adaptive computation and reusable computational structures can improve the efficiency of AI systems without significantly compromising task performance.
Although this is a bold claim, i think it could POTENTIALLY become more mainstream as the use-cases increase for ACS.
The project currently has 4 objectives:
1. Develop the architecture
2. Establish rigourous baselines
3. Measure efficiency
4. Publish findings
I will proceed through implementation, controlled experimentation, failure analysis, and independent evaluation. The research will be guided by falsifiable hypotheses rather than assuming that the proposed architecture is superior since that is a more far fetched goal which is only viable if this whole thing works out.
As mentioned in one of the objectives, success will be determined by measuring performance against established baselines.
I'll measure:
1. Task success and correctness
2. Total model and tool costs
3. completion time
4. Number of failed attempts and retries
5. performance on repeated tasks versus unfamiliar tasks
I am seeking 89000 USD for a 12 month independent research program.
Due to circumstances i dont have a place to live currently so i will try to allocate 2500USD per month to living expenses which adds up to 30000 USD.
9000 USD for Maintaining the Infrastructure and Software
10000 USD contingency
30000 USD for compute, research and model APIs
The initial experiments and prototypes will be expensive but this is necessary to actually establish a fully working model and prove that costs can be cut and intelligence can be made cheaper using ACS
I have a potential collaborator residing in another country who has agreed to help me with ACS if i can provide proof of this being a viable product.
I am at an early stage of research, and I do not yet have a track record of published AI research or demonstrated efficiency improvements from ACS. My current contribution is the initial implementation and the research direction.
The next phase will require more rigorous experimental design, comparisons against existing approaches, and independent evaluation. I intend to make the results available for scrutiny rather than rely on claims about the architecture's potential.
I can promise a working public prototype in around 2-5 weeks after receiving the minimum funding needed for ACS
If the project fails, I expect to produce a documented evaluation identifying which hypotheses were unsupported, which components introduced unnecessary overhead, and under what conditions adaptive computation was ineffective.
That outcome would help prevent further investment in an architecture that does not offer meaningful advantages. It could also identify narrower applications where the approach remains useful.
The project is designed to produce an informative result even if the original hypothesis is wrong.
I have only raised about 2000 USD from my University proffessors to aid my research, due to my personal circumstances ill be starting to fully focus on this full time and have a working model ready by or before April 2027