← Back to Gigs
AI Evaluation
$100 - $120 per hour

LLM Research Scientist (Pre-training & Computer Vision & Adversarial Robustness)

Apply Now → →

SignalGigs may earn a referral fee if you apply through this link.

Description

Research scope

Work on well-scoped but open-ended empirical machine-learning problems across computer vision and language. Projects require end-to-end ownership: preparing data, training models, diagnosing failures, measuring results, and improving performance within strict data, compute, parameter-count, size, or latency budgets. Core responsibilities include:

  • Train image classifiers and generative image models from scratch
  • Fine-tune open-weight language models
  • Improve data efficiency and training efficiency
  • Diagnose optimization, stability, and generalization problems
  • Compress models while preserving accuracy
  • Evaluate results with statistically sound comparisons

Adversarial robustness

Relevant experience may include adversarial training of image classifiers, projected-gradient methods, robustness-focused objectives, and evaluation under standard threat models. Researchers should know how to measure robust accuracy, avoid gradient-masking pitfalls, manage the clean-accuracy versus robustness trade-off, and address robust overfitting. Behavioral robustness in language models is also in scope. Useful work includes shaping multi-turn behavior, resistance to persuasion or sycophancy, calibrated confidence, and determining when a model should accept a correction.

Efficient computer vision

Projects may involve fine-grained image recognition with many similar classes and limited examples. Experience with quantization, pruning, knowledge distillation, and deployment under hard on-device, edge, or embedded constraints is valuable. Researchers should be comfortable selecting architectures and training procedures that preserve accuracy while reducing parameter count, memory use, or inference latency.

Generative image modeling

Relevant backgrounds include training diffusion models, generative adversarial networks, variational autoencoders, or flow-based models from scratch. The work may involve rapid iteration against sample-quality metrics and techniques that produce strong generators under limited training time or small model sizes.

Language-model post-training

Hands-on experience with supervised fine-tuning, preference optimization, reinforcement learning from human or AI feedback, or related post-training methods is useful. Projects may require building datasets through synthetic generation, weak or noisy supervision, rejection sampling, and targeted alignment that changes a specific behavior without degrading general capability.

Multilingual pre-training

Research may cover multilingual or low-resource models trained from scratch. Relevant topics include tokenizer design across scripts and language families, sampling strategies for highly imbalanced language data, and cross-lingual transfer under data constraints.

Additional research areas

Other useful experience includes:

  • Scaling laws and efficient-training research
  • Curriculum learning and data ordering
  • Benchmark design and contamination control
  • Uncertainty estimation and calibration
  • Data augmentation and synthetic data for robustness

Qualifications

Candidates need at least three years of machine-learning research experience; doctoral research counts toward this requirement. Strong hands-on ability with a major deep-learning framework is required. A degree from a highly ranked university, experience at a major technology or AI organization, or an equivalent record of publications and impactful open-source work can demonstrate the expected research level. The engagement is flexible and project based, with compensation described as competitive.

Details

Hiring company

Mercor

Category

AI Evaluation

Location

Remote

Posted

Interested? You'll apply on the hiring company's own site.

Apply Now →

SignalGigs may earn a referral fee if you apply through this link.

Join the job-alert list

Alerts are not being sent yet.

Pay Data

How Much Do AI Jobs Pay?

→
Guide

How to Get Started

→
LLM Research Scientist (Pre-training & Computer Vision & Adversarial Robustness)
$100 - $120 per hour
Apply Now → →