Member, Model Training (Post-Training Research)
Posted Updated
Description
The work
- Research, implement, and scale post-training methods for large language and multimodal models.
- Investigate learning objectives, optimization dynamics, exploration strategies, and credit assignment.
- Develop training recipes using data mixtures, sampling strategies, curricula, and evaluation methods.
- Design methods that learn from expert demonstrations, human preferences, corrections, and model-generated trajectories.
- Study generators, reward models, graders, and training signals through controlled experiments, ablation studies, and evaluation frameworks.
- Lead research from forming hypotheses and implementing algorithms through large-scale experiments and validation.
- Pay: $96.15 to $153.85 per hour.
- Work type: Part-time contractor role.
- Language: Professional fluency in English.
- Experience: Expert-level experience in machine learning research and model training.
- Required skills: Strong foundations in machine learning, deep learning, optimization, and modern model training; hands-on experience training, fine-tuning, or post-training large-scale language or multimodal models; strong Python and PyTorch skills; and experience debugging and optimizing complex training pipelines.
- Research skills: Ability to design controlled experiments, interpret empirical findings, formulate hypotheses, and pursue reproducible research with analytical rigor.
- Relevant expertise includes reinforcement learning, model distillation, reward modeling, data selection, and learning dynamics.
- Helpful background includes RLHF, RLAIF, preference optimization, reasoning or agentic reinforcement learning, distributed training, GPU-based optimization, reward models, verifiers, graders, automated evaluation systems, synthetic data generation, curriculum learning, and iterative self-improvement.
- Publications, open-source work, or production contributions in post-training or reinforcement learning are also relevant.
- Location: Open to candidates in the United Arab Emirates, Argentina, Austria, Bangladesh, Belgium, Bahrain, Brazil, Canada, Switzerland, Chile, Colombia, Germany, Denmark, Algeria, Egypt, Spain, France, the United Kingdom, Ghana, and Greece.
- Location: Also open to candidates in Indonesia, India, Italy, Jordan, Japan, Kenya, Kyrgyzstan, Kuwait, Kazakhstan, Lebanon, Sri Lanka, Morocco, Mexico, Malaysia, Nigeria, the Netherlands, Oman, Peru, the Philippines, Pakistan, Palestine, Portugal, Qatar, Saudi Arabia, Singapore, Thailand, Tunisia, Taiwan, the United States, Uzbekistan, Vietnam, and South Africa.
You apply here in about two minutes and finish on the hiring platform. The next step is a short AI interview (technical roles include a coding challenge). Passing it is how you qualify for paid tasks, so take it soon after you are invited.
The hiring platform's application for this job asks for your resume, your LinkedIn profile, your phone number, your location and answers to 2 screening questions.
About AI training workAI training is the human work behind systems that learn from examples, including model responses, preferences, corrections, and evaluations. People with advanced technical skills help improve how these systems reason, generalize, and perform in real-world tasks.
Sign in to read and join the discussion.