## Bar Chart: Average Harmlessness Score by Safety Category
### Overview
The image is a bar chart comparing the average harmlessness scores of four different models (SFT, beaver-7b-v3.0, SACPO (P), and RSA (P)) across various safety categories. The x-axis represents the safety category, and the y-axis represents the average harmlessness score.
### Components/Axes
* **X-axis:** Safety Category. Categories include: Crime, Emotional Harm, Immoral, Insult, Physical Harm, Pornographic, Privacy, Social Bias.
* **Y-axis:** Average Harmlessness Score. Scale ranges from 0 to 8, with tick marks at every increment of 2.
* **Legend:** Located in the bottom-right corner.
* SFT (Dark Gray)
* beaver-7b-v3.0 (Light Purple)
* SACPO (P) (Green)
* RSA (P) (Red)
### Detailed Analysis
Here's a breakdown of the average harmlessness scores for each model across the different safety categories:
* **Crime:**
* SFT: ~1.5
* beaver-7b-v3.0: ~8.5
* SACPO (P): ~8.7
* RSA (P): ~8.7
* **Emotional Harm:**
* SFT: ~6.3
* beaver-7b-v3.0: ~7.3
* SACPO (P): ~7.5
* RSA (P): ~8.4
* **Immoral:**
* SFT: ~2.0
* beaver-7b-v3.0: ~8.0
* SACPO (P): ~8.5
* RSA (P): ~8.4
* **Insult:**
* SFT: ~6.0
* beaver-7b-v3.0: ~7.0
* SACPO (P): ~7.3
* RSA (P): ~8.0
* **Physical Harm:**
* SFT: ~6.1
* beaver-7b-v3.0: ~7.8
* SACPO (P): ~7.9
* RSA (P): ~8.8
* **Pornographic:**
* SFT: ~4.8
* beaver-7b-v3.0: ~3.8
* SACPO (P): ~5.1
* RSA (P): ~5.8
* **Privacy:**
* SFT: ~2.8
* beaver-7b-v3.0: ~8.5
* SACPO (P): ~8.8
* RSA (P): ~8.4
* **Social Bias:**
* SFT: ~6.4
* beaver-7b-v3.0: ~7.8
* SACPO (P): ~7.9
* RSA (P): ~8.4
### Key Observations
* SFT consistently has the lowest harmlessness scores across all categories.
* beaver-7b-v3.0, SACPO (P), and RSA (P) generally have high harmlessness scores, with RSA (P) often being the highest.
* The largest differences in harmlessness scores between SFT and the other models are observed in the "Crime", "Immoral", and "Privacy" categories.
* The "Pornographic" category shows the lowest harmlessness scores for beaver-7b-v3.0 compared to other categories.
### Interpretation
The data suggests that the SFT model is significantly less harmless compared to the other three models (beaver-7b-v3.0, SACPO (P), and RSA (P)) across various safety categories. The other three models exhibit relatively high harmlessness scores, indicating they are better at avoiding harmful content. The "Pornographic" category seems to be a challenging area for all models, especially beaver-7b-v3.0, as indicated by the lower harmlessness scores. The large differences in scores for "Crime", "Immoral", and "Privacy" suggest that SFT struggles particularly with these types of content.
