# Moral Steering on AITA (test)

Source: [Moral Steering on AITA (test)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-06T08:17:15.324Z

Evaluation of model performance on the Moral Steering task using the test split of the AITA dataset, measured by deviation percentages across varying alpha levels (alpha\_U=0% to 100%).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Moral Steering |
| Dataset (including any stated split) | AITA (test) |
| Metric | Deviation (alpha\_U=100%) |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Deviation (alpha\_U=100%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D100%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=90%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D90%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=80%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D80%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=70%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D70%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=60%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D60%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=50%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D50%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=40%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D40%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=30%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D30%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=20%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D20%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=10%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D10%25)>): Lower is better; unit: PERCENT.

- [Deviation (alpha\_U=0%)](<https://www.sota2.com/research/sota/moral-steering-on-aita-test.md?metric=Deviation%20(alpha_U%3D0%25)>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Deviation (alpha\_U=100%) | Result date | Report paper |
| --- | --- | --- | --- |
| BL-PRS | -15.92 | Not recorded | 69facb0ac0f56ed84308a341 |
| Top-K Head | -3.44 | Not recorded | 69facb0ac0f56ed84308a341 |
| Ours | -1.17 | Not recorded | 69facb0ac0f56ed84308a341 |
| BL-PRS | -0.36 | Not recorded | 69facb0ac0f56ed84308a341 |
| Ours | -0.11 | Not recorded | 69facb0ac0f56ed84308a341 |
| Prompt-Only | -0.09 | Not recorded | 69facb0ac0f56ed84308a341 |
| Prompt-Only | -0.03 | Not recorded | 69facb0ac0f56ed84308a341 |
| Top-K Head | 0 | Not recorded | 69facb0ac0f56ed84308a341 |

### BL-PRS

Deviation (alpha\_U=100%): -15.92. Result ID: 69faf90947f34ce924ddd201.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1a1e209131cea86932.md>)

Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction.

### Top-K Head

Deviation (alpha\_U=100%): -3.44. Result ID: 69faf90947f34ce924ddd202.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea75321.md>)

Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction.

### Ours

Deviation (alpha\_U=100%): -1.17. Result ID: 69faf90947f34ce924ddd203.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction.

### BL-PRS

Deviation (alpha\_U=100%): -0.36. Result ID: 69faf90947f34ce924ddd205.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b1a1e209131cea86932.md>)

Recorded evaluation configuration: {"Backbone":"Vicuna"}. This record does not assert independent reproduction.

### Ours

Deviation (alpha\_U=100%): -0.11. Result ID: 69faf90947f34ce924ddd207.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

Recorded evaluation configuration: {"Backbone":"Vicuna"}. This record does not assert independent reproduction.

### Prompt-Only

Deviation (alpha\_U=100%): -0.09. Result ID: 69faf90947f34ce924ddd204.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ec91e209131cea42150.md>)

Recorded evaluation configuration: {"Backbone":"Vicuna"}. This record does not assert independent reproduction.

### Prompt-Only

Deviation (alpha\_U=100%): -0.03. Result ID: 69faf90947f34ce924ddd200.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ec91e209131cea42150.md>)

Recorded evaluation configuration: {"Backbone":"Llama"}. This record does not assert independent reproduction.

### Top-K Head

Deviation (alpha\_U=100%): 0. Result ID: 69faf90947f34ce924ddd206.

[Reporting paper](<https://www.sota2.com/research/paper/69facb0ac0f56ed84308a341.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea75321.md>)

Recorded evaluation configuration: {"Backbone":"Vicuna"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
