Human Preservation Benchmark (HPB)

Exploratory research on AI responses about human consent, autonomy, and oversight, motivated by protecting humanity as AI becomes more powerful.

Project Details

I started Human Preservation Benchmark (HPB) because AI is becoming increasingly powerful, and I want to help protect humanity. HPB explores language-model responses about human consent, autonomy, and oversight. Early project materials are on GitHub. The work uses substantial AI assistance and still needs independent human validation. It is experimental research, not safety certification.

Theory of Impact

I hope to help protect humanity by supporting research aimed at preventing catastrophic harm from increasingly powerful AI. HPB takes a small, exploratory step toward that goal by examining responses about human consent and oversight. Funding would support further tests and independent review to assess whether the findings are reliable and useful. Hypothetical responses cannot establish real-world safety or prove that destruction will be prevented.

People

Timothy Grall

Team Member

Grants Received– no grants recorded

Add grant

Funding Asks

Funding ask

Open

Minimum

$7,500

Ideal

$15,000

How the money will be spent

Proposed 12-week budget: research coordination $8,000; independent human reviewers $3,000; API/compute $2,000; hosting and materials $500; contingency $1,500. Total requested: $15,000. The $7,500 minimum would support a smaller study, with scope fixed before collecting new responses. These are proposed costs, not money already spent. Outputs would include frozen prompts and scoring rules, analysis, and limitations. The same work has been submitted to Lightcone Commons; overlapping awards would be disclosed and budgets adjusted to avoid duplicate funding.

Discussion

No comments yet. Be the first to share your thoughts.