# Constitution or Collapse? Exploring Constitutional AI with Llama 3-8B

arXiv: https://arxiv.org/abs/2504.04918 · HTML: https://mlsanity.com/paper/2504.04918/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 42 references that are in the map.
- First posted on arXiv 7 Apr 2025.

## Builds on (most cited first, 21 of 21)

- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification](https://mlsanity.com/paper/2107.02314/) (2021)
- [Efficient Multi-Scale 3D CNN with Fully Connected CRF for Accurate Brain Lesion Segmentation](https://mlsanity.com/paper/1603.05959/) (2016)
- [Diffusion Models for Medical Anomaly Detection](https://mlsanity.com/paper/2203.04306/) (2022)
- [MedSegDiff: Medical Image Segmentation with Diffusion Probabilistic Model](https://mlsanity.com/paper/2211.00611/) (2022)
- [TorchIO: A Python library for efficient loading, preprocessing, augmentation and patch-based sampling of medical images in deep learning](https://mlsanity.com/paper/2003.04696/) (2020)
- [Autoencoders for Unsupervised Anomaly Segmentation in Brain MR Images: A Comparative Study](https://mlsanity.com/paper/2004.03271/) (2020)
- [SynthStrip: Skull-Stripping for Any Brain Image](https://mlsanity.com/paper/2203.09974/) (2022)
- [nnU-Net for Brain Tumor Segmentation](https://mlsanity.com/paper/2011.00848/) (2020)
- [Fast Unsupervised Brain Anomaly Detection and Segmentation with Diffusion Models](https://mlsanity.com/paper/2206.03461/) (2022)
- [Unsupervised Anomaly Localization using Variational Auto-Encoders](https://mlsanity.com/paper/1907.02796/) (2019)
- [Mask, Stitch, and Re-Sample: Enhancing Robustness and Generalizability in Anomaly Detection through Automatic Diffusion Models](https://mlsanity.com/paper/2305.19643/) (2023)
- [Rethinking Autoencoders for Medical Anomaly Detection from A Theoretical Perspective](https://mlsanity.com/paper/2403.09303/) (2024)
- [Unsupervised brain lesion segmentation from MRI using a convolutional autoencoder](https://mlsanity.com/paper/1811.09655/) (2018)
- [Parallel Multi-Dimensional LSTM, With Application to Fast Biomedical Volumetric Image Segmentation](https://mlsanity.com/paper/1506.07452/) (2015)
- [Modality Cycles with Masked Conditional Diffusion for Unsupervised Anomaly Segmentation in MRI](https://mlsanity.com/paper/2308.16150/) (2023)
- [IterMask2: Iterative Unsupervised Anomaly Segmentation via Spatial and Frequency Masking for Brain Lesions in MRI](https://mlsanity.com/paper/2406.02422/) (2024)
- [Feasibility and benefits of joint learning from MRI databases with different brain diseases and modalities for segmentation](https://mlsanity.com/paper/2405.18511/) (2024)
- [Deepbet: Fast brain extraction of T1-weighted MRI using Convolutional Neural Networks](https://mlsanity.com/paper/2308.07003/) (2023)
- [Deformation-Recovery Diffusion Model (DRDM): Instance Deformation for Image Manipulation and Synthesis](https://mlsanity.com/paper/2407.07295/) (2024)
- [An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training](https://mlsanity.com/paper/2206.03359/) (2022)

## Built on by (most cited first, 4 of 4)

- [A Lightweight Explainable Guardrail for Prompt Safety](https://mlsanity.com/paper/2602.15853/) (2026)
- [Beyond Creed: A Non-Identity Safety Condition A Strong Empirical Alternative to Identity Framing in Low-Data LoRA Fine-Tuning](https://mlsanity.com/paper/2603.14723/) (2026)
- [Ethical Risks in Deploying Large Language Models: An Evaluation of Medical Ethics Jailbreaking](https://mlsanity.com/paper/2601.12652/) (2026)
- [Evaluating Bounded Autonomy in Regulated Agentic AI: A Diagnostic Harness with Constitutional Rewards, Escalation Labels, and Runtime Governance](https://mlsanity.com/paper/2609.37501/) (2026)

## Abstract (arXiv)

As language models continue to grow larger, the cost of acquiring high-quality training data has increased significantly. Collecting human feedback is both expensive and time-consuming, and manual labels can be noisy, leading to an imbalance between helpfulness and harmfulness. Constitutional AI, introduced by Anthropic in December 2022, uses AI to provide feedback to another AI, greatly reducing the need for human labeling. However, the original implementation was designed for a model with around 52 billion parameters, and there is limited information on how well Constitutional AI performs with smaller models, such as LLaMA 3-8B. In this paper, we replicated the Constitutional AI workflow using the smaller LLaMA 3-8B model. Our results show that Constitutional AI can effectively increase the harmlessness of the model, reducing the Attack Success Rate in MT-Bench by 40.8%. However, similar to the original study, increasing harmlessness comes at the cost of helpfulness. The helpfulness metrics, which are an average of the Turn 1 and Turn 2 scores, dropped by 9.8% compared to the baseline. Additionally, we observed clear signs of model collapse in the final DPO-CAI model, indicating that smaller models may struggle with self-improvement due to insufficient output quality, making effective fine-tuning more challenging. Our study suggests that, like reasoning and math ability, self-improvement is an emergent property.