Academic Review NeurIPS 2024. This year at the NeurIPS conference we will analyze and discuss with academics, and forward-thinking research that could aid in AGI progress. We believe the brain is doing something unique, and to create general intelligence we must look to nature. The week-long conference covers many topics in Neuroscience and computation. We will attend lectures and talk with presenters about how their research could help progress in the AGI space as it relates to the Hyperon framework. Companies like Cortical Labs and Arc Prize will be attending with updates on their progress. We attended the 2024 symposium ‘The Generative Mind: Biological and Artificial Intelligence’ which presented that slight changes in algorithms that can formulate calculations to run like neurons!¹ The Neurips conference will be an expanded version that will discuss cutting-edge discoveries and research that will support AGI. An example lecture at Neurips - Causality for Large Language Models.² ¹ https://arxiv.org/abs/2309.08045 ² https://arxiv.org/abs/2410.15319

Post conference we will curate a list of research papers, academics, ideas, and interesting attendees all adding to progress towards AGI. This will include potential companies working on AGI, researchers, current and alternative approaches to AGI. Contact details and research affiliations, if approved by persons, will be publicly available in the database/website. The data will also include how these approaches can be implemented and/or assist in the Hyperon framework.

$25,000 USD

An online database/website with NeurIPS data publically available to access and share information. Users will be able to submit additional information, comment and become a hub for knowledge. Milestone 1 will also be the beta version of the website as our information grows so will the website

Milestone 1 will present cutting-edge research and ideas in the space of biological intelligence, artificial intelligence, and forward-thinking ideas in current AGI approaches. Milestone 1 is in place to absorb the latest ideas and solutions from neuroscience. We think the brain has insights we can gain to help us build better AI systems. With that fresh in mind we will do a full review of the Singularity net ecosystem and its architecture; MeTTa, MeTTa compilers, and Hyperon. A detailed review and evaluation of components. We will speak with engineers and the community to understand current projects, limitations, and future vision. What tools exist, or missing? What hardware is currently being used, and what can be tested? What hardware would be best to improve capabilities when looking at specs, power, and output? The research will include the pieces needed to build a benchmark test for hardware and asses capabilties.

The website/database will be updated which includes findings, projects/Github, and links from the community. It will help us better understand the entire stack so that we can determine what type of hardware would accelerate the development of AGI. An outline of a sandbox testing environment will be presented, with input from the community and Hyperon engineers.

$15,000 USD

We understand the goals and architecture of Singularity Net, and present our benchmark proposal into our database/website for external review. This may change as we begin our research, and we may choose to pick a particular aspect to work on and benchmark. The benchmark will be opensource.

Outline and research companies that could contribute to creating AGI and how they could fit into the Hyperon ecosystem. Could pieces be improved with distributed software running on Nvidia Jetsons or can Tenstorrents Grayskull outcompute? power to compute ratio? Can reversible computers be utilized? What can Thermodynamic chips help with? Can a brain organoid compute? Interviewing and assessing hardware and software companies currently available on the market, and companies building. If available, get specs on devices, get updates, and share publicly if approved. This includes hardware and software technologies to help run algorithms, distribute compute, power abilities, and novel software. The list below is a start, some may not want to discuss publicly and will be adjusted as research expands. Some metric assessments include technical specs, power consumption, and ability to compute in decentralized environments (e.g.. EXO labs). GSI Technolgy (available) - HW TensTorrent (available) - HW MatX (building) - HW Recogni (available) - HW SiMa.ai (available?) - HW Celestial Ai (unknown) - HW Lightmater - (unknown) - HW Blumind - (unknown) - HW Taalas - (unknown) - HW Innatera (unknown) - HW Eliyan (available) - HW Vaire computing (building) - HW Extropic (building) - HW Cortical Labs (building) Wetware FinalSPark (building) Wetware TinyBox/Comma Ai (available) - Hardware/software Keen Technologies - Unknown Exabits - Distributed compute EXO Labs - Distributed compute

The same methods from Milestone 1 will be implemented into the website/database. Hardware and software research will be uploaded and publicly available. Forward-thinking statements will also be included, with a focus on current technology. The database will also include where the research can fit into Hyperons stack. The database/website format will be chosen based on best use. It will list out the challenges that exist and potential solutions.

$20,000 USD

An aggregate list of hardware/software companies capable of progress towards AGI, easily accessible to the public, with detailed information on hardware/software specs. The list will continue to grow after initial research is complete. As companies come out of stealth and others are discovered, the data will help developers find potential solutions to the challenging AGI stack to be created.

With a full understanding of new ideas, software architecture, and hardware landscape, a small beta test will be run using the Hyperon ecosystem. This will include hardware available today, for example, Tenstorrent, Nvidia, and GSI. Post Milestone 2 the benchmark will be developed and tested. This will be put into a GitHub repository that will be open source that anyone can run. The idea behind this is to develop something that the AI community, students, or hobbyists can play with and understand. Milestone 4 may pivot to a different approach after researching previous milestones, which could include running a decentralized cluster of compute chips for example. Dependent on previous research Milestone 4 will be updated to reflect the most beneficial hardware testing. This contributes knowledge, understanding, and targets limitations that will benefit the AGI ecosystem. Before completing previous milestones we plan to test operations on these devices: Tenstorrent GraySkull Nvidia Jetson Nvidia RTX 4090 GSI Technolgy* *This card may be out of price range, but if research heads in this direction direct contact for a demo or cloud access will be requested.

A video or blog post added to the website/database listed above documenting the process, challenges, and code. A benchmark of stats relevant to Hyperon/MeTTA. We will also raffle away the hardware we purchased if the community approves. We will create a prize, puzzle or bounty to give away one or more of the pieces of hardware purchased.

$20,000 USD

A publicly available repository for users to access and test our demo setup. The community approves and continues to grow/fork our work to create useful testing applications, or inspires creative alternative solutions based on our research. After Milestone 4 we have a database/website available to the public with detailed information showcasing the latest in Neuroscience’s approach to AGI, in both software and hardware. The Hyperon ecosystem is referenced with links to the Neuroscience research including a detailed description of the latest hardware developments. The hardware details include specs on abilities, current availability, and timelines for promising deep-tech hardware under development. All of this combined will provide the community with a great resource. Our goal is to finish this deep discovery to engage with the community, discover potential hardware paths, engage with startups building, and/or create their own papers with novel methods to forge the future of AGI. AGI won't come from statistical next word prediction.