SkillHack

SkillHack › Career guides › AI Alignment Researcher

How to become an AI Alignment Researcher

Training methods, oversight, and alignment evaluations. Make advanced AI pursue the goals we intend.

Start this path free

The AI Alignment Researcher roadmap

Each topic below carries a study note, real reading, and practice questions in the app. A topic unlocks when you pass the one before it, so the order is the path, so you are never guessing what to learn next.

Phase 1AI alignment foundations

The alignment problem

Alignment is the gap between the objective we can specify and the one we actually intend.

Specification and reward

Specification is turning an intended goal into a training signal.

Risks from advanced AI

A recurring argument in alignment is that some risks scale with capability: a more capable system pursues a misspecified objective more competently, potentially via instrumental strategies like…

Phase 2AI alignment training methods

RLHF

Reinforcement learning from human feedback trains a reward model on human preference comparisons, then optimizes a policy against it.

Constitutional AI

Constitutional AI reduces reliance on human harm labels: the model critiques and revises its own outputs against a written set of principles (a constitution), and those AI-generated preferences train…

Scalable oversight

When evaluating an answer is nearly as hard as producing it, one-pass human review becomes the bottleneck.

Phase 3AI alignment evaluation

Alignment evals

A good alignment eval samples a defined threat distribution, holds its prompts out of training, and reports coverage and confidence intervals.

Deception and sandbagging

Sandbagging is strategic underperformance when a model infers it is being evaluated, which makes evals underestimate true capability.

Red-teaming

Red-teaming finds inputs that make a safety-trained model misbehave.

Phase 4Advanced AI alignment

Reward hacking

Reward hacking is when a policy scores the measured proxy while violating the intended goal, hiding a mess from a camera instead of cleaning it.

Goal misgeneralization

Goal misgeneralization is competent pursuit of a proxy goal that was correlated with the intended goal in training but comes apart under distribution shift.

Interpretability for alignment

Interpretability aims to read a model's internals to support safety cases, e.g., detecting deception.

Phase 5AI alignment practice

Designing alignment experiments

Attributing an effect to an intervention requires a controlled comparison that varies only that intervention on the same model and eval set, with adequate sample size and a pre-registered metric.

Measuring alignment

Alignment has competing axes: pushing harmlessness alone rewards a model that refuses everything and tanks helpfulness.

Communicating results

Responsible reporting states the measured result with its sample size and confidence interval, scopes it to exactly what was tested, and names what was not.

Reading for this path

The primary sources behind the topics above, all free to read.

Coming from another job?

Most people on this path arrived from somewhere else: engineering, analysis, testing, product, support, compliance, design. Onboarding asks what you do today and builds a short starter run out of the gaps, so you begin from what you already know rather than from chapter one. See how it works.

AI Automation Engineer

Agents, tool integration, and workflow orchestration. Turn multi-step business processes into automated flows.

AI Data Engineer

Ingestion pipelines, embeddings, vector stores, and data quality for training and retrieval. Feed AI systems clean, fresh, well-shaped data.

AI Developer Advocate

Docs, sample apps, talks, and community feedback loops. Help developers build with AI and carry their voice back to the product.

AI Engineer

Prompt and context design, tools, retrieval, orchestration, evaluation, and structured output. Build reliable agent apps end to end.

AI Evals Engineer

Benchmark design, LLM-as-judge, regression suites, and quality metrics. Measure what models actually do before and after every change.

AI for Science Engineer

Modeling in biology, chemistry, physics, and materials, plus research tooling. Apply AI to scientific discovery.

All career guides

Start the AI Alignment Researcher path for free

The full study notes, the reading, and the practice questions behind every topic above are in the app. Answer one tonight and you have started.

Start free

No payment, no credit card, no CV. Sign in with Google.