Roadmap
AI Researcher
Becoming a AI Researcher means building the skills below in roughly this order. The current median pay for AI Researcher roles on Kairos is $274,500 (369 disclosed listings). 17 of these skills appear in live AI Researcher listings right now.
Demand and pay updated 4 hours ago.
Math & programming foundations
0-6 months- Core
Research demands more math than applied ML - comfort with proofs, gradients, and probabilistic reasoning.
- In demand · 48%Core
Fluent Python with NumPy for fast, correct experiment code.
- In demand · 71%Core
The why behind the methods - bias/variance, generalization, and optimization.
Deep learning mastery
6-12 months- In demand · 25%Core
Architectures, training dynamics, and the transformer in depth - from the math to the code.
- In demand · 35%Core
Research-grade frameworks; JAX and CUDA matter for performance-critical work.
- Recommended
Re-implementing results from scratch is the fastest way to research competence.
Choose a specialization
12+ months- In demand · 40%Optional
Pretraining, alignment, and evaluation of large language models.
- In demand · 25%Optional
Policy learning and RLHF - central to modern alignment and agents.
- In demand · 19%Optional
Diffusion models, multimodal architectures, and perception.
Publish & scale
Ongoing- Recommended
Track arXiv, write up results, and submit to venues like NeurIPS, ICML, and ACL.
- In demand · 20%Optional
Train at scale across many GPUs - the practical bottleneck for frontier work.
Grounded in
- Stanford CS229 - Machine Learning ↗ - Graduate ML foundations (theory-forward).
- Stanford CS224N - NLP with Deep Learning ↗
- DeepLearning.AI - Deep Learning Specialization ↗
- fast.ai - Practical Deep Learning ↗
See also
Questions
- How do I become a AI Researcher?
- Work through the roadmap in order: start with the Core foundations, then layer on the Recommended and Optional skills. The path toward research scientist and research engineer roles at AI labs: strong math and programming foundations, deep learning mastery, a specialization (LLMs, RL, vision, or systems), and the ability to read, reproduce, and publish papers.
- Is this roadmap based on real sources?
- Yes. It is grounded in Stanford CS229 - Machine Learning, Stanford CS224N - NLP with Deep Learning, DeepLearning.AI - Deep Learning Specialization, fast.ai - Practical Deep Learning. Skill demand and pay are measured live from AI job listings on Kairos.