You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$200,000 left to reach the funding goal.
TLDR: Some of my research thoughts and work can be found here. But I think my core interests are (1) Searching for universal alignment techniques, things that transfer across paradigms. I think this will be important when/if we move away from transformers, and to also help us glean a universal understanding of cognition which can fuel further research. (2) A better understanding of agency. If we can understand what kind of optimal agent we will get from a certain training method, reward function, optimization pressure, and environment, then we can potentially run a small scale simulation of a lot of said agents and be able to “pluck out” the optimal ones for humans. (3) Cover more ground! If you’re uncertain where the optimal research direction leads, try a lot and see what yields fruit. We can also use AI capabilities to our benefit. That’s why I created v1 of Alignment Commons, where I am creating tests to run on hundreds of research papers (feel free to connect your own agent to the workspace, just hand it to Claude, and it’ll do the rest). I’m doing my own research, but am bottlenecked by compute, and think I could do far more if I was able to focus on this full time.
Details
Research testbed: https://www.aisafetyfund.org/research
Alignment Commons: https://alignment-commons-site.vercel.app/
Research Projects: https://alignment-commons-site.vercel.app/workspace#projects
Alignment Commons Agent workspace (add your own agent and message me): https://alignment-commons-site.vercel.app/workspace
Other projects: https://www.aisafetyfund.org/sponsored-projects
The compute budget (listed below) would go toward scaling the testbed to handle larger experiments. The salary would go toward me scaling this, other projects, managing the AISR fund, and my own research I'm conducting at the Fund (linked below).
Theory of Impact
Research Outlook and preliminary results: https://docs.google.com/document/d/1ToYtvVI-Dvn2ZmlVIM_zRDFDz3357wJ-U9nfidNXSD0/edit?usp=sharing
Further thinking: https://aisafetyfund.substack.com/p/stop-saying-ai-alignment
Budget - Compute
OpenAI Pro: $200
Anthropic Ultra: $200
Runpod: $100-$2,000
Ollama: $100-$500
Anthropic API: $20-$200
Kaggle: ~$200
Budget - Other
One year salary: $80,000-$200,000 (if high-enough, will move to sf)
Travel, events: $2000
About me
Website: https://jaesonbooker.com/
AISR Fund: https://www.aisafetyfund.org/
There are no bids on this project.