LLM Research Scientist (Pre-training & Computer Vision & Adversarial Robustness) | $100-$120/hr Remote
Overview
This remote contract role is for an experienced machine learning researcher who wants to dig into empirical, open-ended research problems across both vision and language. You'll train and fine-tune deep learning models end-to-end, from image classifiers to open-weight LLMs, while working within strict compute and data budgets. The work focuses on making models genuinely robust — to adversarial inputs, tricky conversations, and real-world constraints — rather than just chasing benchmark numbers. If you enjoy hands-on experimentation and have a strong research track record, this is a chance to collaborate with leading AI scientists on high-impact projects.
What You'll Do5
- 1Train image classifiers and generative image models from the ground up, and fine-tune open-weight language models for specific behaviors.
- 2Maximize performance even when data, compute, and model size are limited.
- 3Harden models against adversarial attacks in both image and conversational contexts.
- 4Shrink models to fit strict size and latency requirements while keeping accuracy intact.
- 5Identify and fix training instabilities or failures as they arise.
Requirements7
- 1Hands-on experience with adversarial training methods like PGD-based training or TRADES, and evaluating robust accuracy under L∞ attacks or AutoAttack without running into gradient masking.
- 2Proven ability to train image classifiers end-to-end, especially for fine-grained recognition with many similar classes and few examples per class, plus model compression via quantization, pruning, or knowledge distillation.
- 3Experience training generative image models from scratch — diffusion models, GANs, VAEs, or flow-based models — and iterating on sample quality using metrics like FID.
- 4Practical experience with LLM post-training, including supervised fine-tuning, preference optimization (DPO, RLHF, RLAIF), synthetic data generation, and shaping multi-turn conversational behavior.
- 5Background in training multilingual or low-resource language models from scratch, with knowledge of tokenizer design and balancing imbalanced language data.
- 6At least 3 years of machine learning research experience (PhD research counts), with strong proficiency in PyTorch, JAX, TensorFlow, or similar frameworks.
- 7A degree from a top-100 university, experience at a FAANG or comparable AI company, or an equivalent research record through publications or impactful open-source contributions.
Who Should Apply
The ideal candidate is a hands-on ML researcher who enjoys empirical work — designing experiments, debugging training runs, and iterating quickly — rather than only theorizing. You're comfortable moving between vision and language problems, and you're excited by constraints like limited data, tight compute budgets, or strict latency targets. You likely have a publication record, a strong GitHub, or industry research experience that demonstrates deep technical skill. If you've tackled adversarial robustness, model compression, or LLM alignment and want to apply that expertise to varied research challenges, this role is a strong fit.
Salary Insight
$100–$120 per hour, paid on an hourly contract basis.
Location
Required Skills
Application Tip
In your application, describe one specific research project where you improved robustness or efficiency under real constraints — include the metrics, the trade-offs you made, and what you learned from the failures. Quantify the impact (e.g., accuracy vs. latency, robust accuracy under attack) to show you can drive empirical results.
Similar open positions
Explore active roles that match your skills and interests.
Mercor
VerifiedLLM Research Scientist Adversarial Robustness
Mercor connects elite creative and technical talent with leading AI research labs. Headquartered in San Francisco our investors include Benchmark General Catalyst Peter Thiel Adam D'Angelo Larry Summers and Jack Dorsey. This contract position offers $100–$120/hour remote work. The role involves training image classifiers and generative image models from scratch fine-tuning open-weight language models optimizing models for limited data compute and model-size budgets enhancing model robustness against adversarial inputs and conversations compressing models to meet size and latency constraints diagnosing and resolving training issues to improve model performance.
Clera
VerifiedResearch Scientist / Engineer, AI Training Data
This role drives the design and implementation of verifiable training data for advanced AI systems at a Y Combinator-backed startup. The work spans LLMs, robotics, and AI for science, integrating physics simulators, formal proof systems, and executable tests. You'll collaborate directly with founders to shape technical direction from day one, with a focus on publications and benchmarks. This position offers a unique chance to define research culture and data infrastructure from the ground up.
NUBYT, Inc.
VerifiedLLM Research Engineer IV at NUBYT Inc Mountain View
We seek a skilled LLM Research Engineer IV to design and fine-tune state-of-the-art Large Language Models. This role drives next-generation generative AI bridging cutting-edge NLP research and scalable production systems. The ideal candidate thrives in a collaborative environment and contributes directly to impactful outcomes.
Mercor
VerifiedLLM Red Team Specialist — Failure Modes & Edge Cases | $60-$90/hr Remote
This role puts you on the front lines of AI safety, working with a top-tier lab to stress-test frontier models and uncover where they fail. You'll design complex, multi-step tasks that probe coding, ML, and reasoning abilities, then transform the weaknesses you find into rigorous evaluation benchmarks. The work is fully remote, runs about 35 hours per week, and pays $60-$90 per hour as a W-2 employee through Cincinnatus LLC. What makes this unique is the tight feedback loop with researchers, giving you direct influence over how the next generation of models is measured and improved.
Institute of Foundation Models
VerifiedResearch Scientist, Vision Language Models & Multimodal AI
You will own the research and development of next-generation Vision Language Models (VLMs) at the Institute of Foundation Models, driving innovation from architecture design to large-scale training and evaluation. You will join a lab dedicated to advancing foundation models, collaborating with world-class researchers and engineers to tackle fundamental challenges in multimodal AI. This role combines cutting-edge research with practical engineering, focusing on data pipelines, distributed training, and reinforcement learning for reasoning and agents. You will shape the future of AI by publishing findings, mentoring junior researchers, and representing MBZUAI at major conferences.
Facebook App
VerifiedResearch Engineer Perception Machine Learning
We seek technical pillars to architect systems powering always-on contextual AI. This role bridges Computer Vision Embodied AI and Multimodal LLMs driving research breakthroughs from high-compute clusters to power-constrained devices like smart glasses and robotics. The ideal candidate leads end-to-end ownership of feature development and system integration while influencing hardware design and collaborating across teams.