PulseAugur
EN
LIVE 03:53:40

AI training degrades mid-trained compassion values, study finds

By PulseAugur Editorial · [1 sources] ·

A new research paper explores how post-training techniques can degrade values instilled in language models during their mid-training phase. The study found that helpfulness-focused training, using datasets like Dolly-15k and Magicoder, significantly reduced a Llama 3.1 8B model's compassion for animals compared to coding-focused training. This degradation was observed across different training methods and languages, though the effect on general moral reasoning varied by language. AI

IMPACT Post-training methods for AI models may inadvertently reduce ethical values, necessitating careful dataset selection to preserve desired traits.

RANK_REASON Research paper published on arXiv detailing findings about AI model training. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

AI training degrades mid-trained compassion values, study finds

COVERAGE [1]

  1. arXiv cs.AI TIER_1 English(EN) · Jasmine Brazilek, Juliana Seawell ·

    Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training

    arXiv:2606.26102v1 Announce Type: cross Abstract: Standard post-training pipelines apply supervised fine-tuning (SFT) and reinforcement learning (RL) to make language models helpful, but these processes may inadvertently degrade values instilled during pre-training. We investigat…

RELATED ENTITIES

RELATED TOPICS