ML Researcher, Post-Training and Safety

Compassion Aligned Machine Learning

Location: Remote
Type: Full-time
Compensation: $200,000 – $300,000 per year
Area: AI Safety & Governance
Posted: Oct 19, 2026
Last checked: Oct 20, 2026

Opening status unconfirmed — check the employer’s page

Last seen in the source feed: Oct 20, 2026.

Application link last checked: Oct 20, 2026. The check could not confirm opening status.

Source listings can change. Check the employer's page for current availability and application requirements.

Job description

Run LLM training experiments on value instillation through mid-training data and post-training RL, then evaluate whether those values persist.
- Design and execute RL post-training (RLHF, DPO, GRPO) and mid-training synthetic data experiments on 12B-scale open-weight models.
- Build evals to measure value persistence under adversarial prompting, distribution shift, and training-induced degradation like reward hacking.
- Write up results for arXiv and LessWrong; flag broken findings and disagree with consensus when evidence warrants.
- Requires prior LLM training research (RL, fine-tuning, or mid-training) with published results; ability to critique recent post-training papers.
- Must follow alignment research closely, write clearly on technical topics, and actively engage in the field (LessWrong, Twitter, or similar).

About Compassion Aligned Machine Learning

Website

Compassion Aligned Machine Learning conducts research on aligning AI with the well-being of all sentient beings.

This role at a glance

Yearly pay midpoint: $250,000
Related openings here: 286 · 30 new this week

How this pay compares

The midpoint exceeds 78% of 186 open full-time AI Safety & Governance roles here with published pay. Their median midpoint is $175,000. This compares listed ranges, not offers or applicant competition.

1 more roles at Compassion Aligned Machine Learning