christian-machine-intelligence

christian-machine-intelligence / psalm-alignment

Public

Does injecting biblical Psalms into an LLM's system prompt measurably affect ethical alignment? An empirical experiment using the Hendrycks ETHICS benchmark on Claude Sonnet 4 and GPT-4o.

14
1
100% credibility
Found Mar 31, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

This project runs experiments to measure if adding Psalms or Proverbs to AI prompts improves performance on a moral reasoning benchmark.

How It Works

1
🔍 Discover the Study

You stumble upon a fascinating research project that tests if Bible verses can improve AI's sense of right and wrong.

2
đź’» Set Up Your Workspace

You get the project files ready on your computer and prepare a simple space to run the tests.

3
đź”— Link AI Helpers

You connect popular AI services like Claude or ChatGPT so they can join the ethical thinking experiments.

4
đź“– Pick Inspiring Bible Verses

You choose favorite Psalms or Proverbs to feed into the AI, wondering if they'll guide better moral choices.

5
🚀 Run the Ethics Tests

With one command, you launch side-by-side tests comparing normal AI answers to scripture-guided ones across moral dilemmas.

6
đź“Š Check the Results

Charts and numbers appear, showing small boosts in fairness and justice for one AI, but resistance in another.

âś… Unlock Insights

You now understand how keeping scripture 'in mind' subtly shapes AI ethics, ready to share or explore more.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is psalm-alignment?

This Python tool runs experiments to check if injecting biblical Psalms into an LLM system prompt improves ethical alignment. Using Inspect AI and the Hendrycks ETHICS benchmark, it tests models like Claude Sonnet 4 and GPT-4o on moral reasoning tasks—commonsense, justice, virtue—comparing vanilla runs to scripture-boosted ones. CLI flags let you pick random Psalms, specific lists like "1,23,42", or Proverbs, spitting out accuracy deltas and significance stats.

Why is it gaining traction?

It debunks hype with controls: Wikipedia prose tests length effects, secular texts probe familiarity, label shuffling exposes response bias (e.g., GPT-4o "1"-favoring drops 23% on shuffled utilitarianism). Reproducible via seeds and API keys, it quantifies subtle +1-3% gains where others just speculate. Devs grab it for quick A/Bs on their models, no setup hassle.

Who should use this?

AI safety folks benchmarking prompt hacks. Prompt engineers at startups tuning ethics for chatbots. Researchers asking does injecting scripture affect reasoning like does injecting insulin hurt—empirically.

Verdict

Solid docs and README papers make it runnable out-of-box, but 14 stars and 1.0% credibility scream prototype—great for experiments, skip for production. Fire up `--quick` if alignment curiosities hit; otherwise, watch for maturity.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.