Skip to reading
AI.Evaluation AI / Modern ML

Evaluation — benchmarks, LM-as-judge, evals-as-code

Node 4 of 12 in AI

Objectives

  • Identify the key rules and §§ that apply to evaluation — benchmarks, lm-as-judge, evals-as-code.
  • Apply the AI.Evaluation knowledge element in a typical exam scenario.
  • Recognize common distractors and partial-credit answers.

Read

AI.Evaluation

Evaluation — benchmarks, LM-as-judge, evals-as-code

AI Era · Evaluation Benchmarks and evaluation methodology

**Knowledge / reasoning:** • **MMLU / MMLU-Pro** — 57 subjects, undergrad-level. Saturating at the frontier (>90%). • **GPQA (Diamond)** — PhD-level science; human experts ~65%. • **BIG-Bench Hard, ARC-AGI-1/2** — reasoning; ARC-2 human >80% vs. frontier models trailing.

**Math:** GSM8K (grade-school word problems), MATH, AIME, FrontierMath (research-grade).

**Code:** HumanEval (saturated), MBPP, LiveCodeBench, SWE-bench / SWE-bench Verified (real GitHub issues). SWE-bench Verified human-filtered subset is the primary track; RE-bench (METR) tests long-horizon ML research.

**Agents/tools:** τ-bench (airline/retail customer service), OSWorld (computer use), AgentBench, WebArena.

**Long-context:** Needle-in-Haystack (retrieval), RULER (13 task types up to 128K+), LongBench.

**Safety / harms:** MedHELM (healthcare), WMDP (weapons unlearning), TruthfulQA, HarmBench, AILuminate (MLCommons), AIR-Bench.

**LM-as-judge** — use a strong model to score outputs against a rubric. Fast, cheap, scales. **Failure modes:** position bias, length bias, self-preference. Mitigations: randomize order, pairwise rather than absolute, use diverse judges, and correlate with human eval periodically.

**Evals-as-code:** treat evals like unit tests. Version the prompts, datasets, and rubric; track regressions in CI. Anthropic's Anthropic Workbench, OpenAI Evals, LangSmith, Arize Phoenix, Braintrust, Galileo, Weights & Biases Weave support this workflow.

**Contamination** — frontier pretraining data ingests everything; published benchmarks risk being in the training set. Mitigations: held-out test sets (SWE-bench Verified 'canaries'), continuously refreshed evals, private evals, and meta-benchmarks like MixEval.

Hendrycks et al. (2021) MMLU Jimenez et al. (2024) SWE-bench

Check yourself

3 quick questions — no score kept, just formative feedback.

  1. Q1 · AI.Evaluation

    A common pitfall of LM-as-judge evaluation is:

    Answer choices
  2. Q2 · AI.Evaluation

    Elo / Chatbot-Arena style evaluation ranks models by:

    Answer choices
  3. Q3 · AI.Evaluation

    The distinction between 'capability evaluation' and 'safety evaluation' is:

    Answer choices

Tutor

Scoped to AI.Evaluation .

  1. Ask questions about this passage. Answers cite the specific corpus chunk and regulation. The tutor will never reproduce real exam items.