LLM Research Scientist – Pre-training, Computer Vision, Adversarial Robustness

This is a fully remote position, open to applicants in United States.

📋 Description

• Develop image classifiers and generative image models from the ground up.

• Fine-tune pre-existing open-weight language models.

• Engage in empirical and exploratory machine learning research.

• Enhance models while operating within constraints of limited data, computational resources, and model size.

• Increase model robustness against adversarial inputs and challenging conversations.

• Compress models to adhere to size and latency requirements without compromising accuracy.

• Identify and resolve training challenges.

• Collaborate with top AI researchers on impactful and demanding projects.

• Participate in initiatives aimed at training and improving AI systems.

⛳️ Requirements

• Over 3 years of experience in machine learning research; PhD research can be included.

• Proficient in PyTorch, JAX, TensorFlow, or other similar ML frameworks.

• Degree from a top-100 university, experience with a FAANG or similar AI organization, or a strong research background demonstrated through publications or significant open-source contributions.

• Skilled in adversarial training of image classifiers and robust accuracy evaluation, as well as understanding robustness–accuracy trade-offs.

• Knowledgeable in model compression techniques such as quantization, pruning, and knowledge distillation.

• Experienced in deploying models with strict size or latency constraints.

• Familiar with training diffusion models, GANs, VAEs, or flow-based generative models from inception.

• Competent with FID metrics and training efficiency methods for generative models.

• Experienced in supervised fine-tuning and preference optimization (DPO, RLHF, RLAIF) of open-weight language models.

• Proficient in dataset creation through synthetic generation, noisy or weak supervision, and rejection sampling.

• Experienced in shaping conversational behavior across multiple interactions.

• Knowledgeable in alignment-style fine-tuning while maintaining general capabilities.

• Skilled in training multilingual or low-resource language models from scratch.

• Experienced in designing tokenizers for diverse scripts and languages.

• Capable of balancing uneven data across languages.

• H1-B and STEM OPT candidates are not supported.

🏝️ Benefits

• Flexible, project-oriented work schedule.

• Competitive salary offerings.

• Fully remote position.

• Ability to work on your own timetable.

• Weekly payments processed through Stripe or Wise.

• Project timelines can be adjusted based on requirements and performance evaluations.

• Collaborate with leading AI researchers.

• Reasonable accommodations provided upon request.