// ⛨ Algorithms Atlas

Safety, Evaluation & Security

Adversarial examples, jailbreaks, prompt injection, red teaming, watermarking, differential privacy and scalable oversight.

See all 11 ↓ All areas →

8 core3 extended2014–2024

TIMELINE

2014 Adversarial Ex2016 Membership Inf2016 Differentially2018 AI Safety via 2022 Automated Red 2023 GCG Universal 2023 Indirect Promp2023 LLM-as-a-Judge2023 LLM Watermarki2023 Weak-to-Strong2024 Sleeper Agents

01Core algorithms

Learn these first. Sorted by year.

Adversarial Examples (FGSM) Imperceptible input perturbations that fool models. 2014core Membership Inference Attacks Tells whether a given record was in a model’s training set – a measure of privacy leakage. 2016core Automated Red Teaming Uses one LM to generate test cases that attack another. 2022core GCG Universal Jailbreak Attack Automatically finds gibberish suffixes that jailbreak aligned models. 2023core Indirect Prompt Injection Hides malicious instructions in content the model reads. 2023core LLM-as-a-Judge Strong LLMs grade or compare other models' answers. 2023core LLM Watermarking Biases generation toward a hidden green list so AI text is detectable. 2023core Weak-to-Strong Generalization Supervising strong models with weak-model labels as an analogy for superalignment. 2023core

02Extended

Variants and follow-ups worth knowing once the core is clear.

Differentially Private SGD (DP-SGD) Clips per-example gradients and adds noise to protect individuals. 2016ext AI Safety via Debate Two AIs debate and a human judges, aiding oversight of hard questions. 2018ext Sleeper Agents (Backdoors that Persist) Backdoored models keep hidden behaviors even after safety training. 2024ext

03Course units that teach these

U81 RAG Engineering & Evaluation Chunking, query rewriting, HyDE, GraphRAG, citations and evaluation metrics. 8 h4 algorithms3 paperslesson ready U83 AI Engineering: Evals, Cost & Shipping Your own eval sets, LLM judges, observability, model gateways, cost control and product design. 8 h3 algorithms1 papers U90 Evaluation & Benchmarks Major benchmarks, arenas, data contamination and the limits of capability evaluation. 6 h2 algorithms2 paperslesson ready U92 Security: Jailbreaks, Injection & Privacy Adversarial examples, GCG, prompt injection, red-teaming, watermarking, differential privacy and backdoors. 6 h8 algorithms4 paperslesson ready U93 Frontier Risk, Scalable Oversight & Governance Weak-to-strong generalisation, debate, Constitutional AI, risk frameworks and regulation (EU AI Act, NIST AI RMF). 6 h4 algorithms3 papers