// ⌖ Interpretability

Logit Lens & Probing

Decodes intermediate layers into tokens, or probes them with small classifiers.

Logit Lens 与探针 2020 Intermediate core

IN PLAIN WORDS

The logit lens multiplies each layer's hidden state by the unembedding to see the model's interim guess; probes train linear classifiers to test what a layer encodes. Entry-level tools for model internals.

Used in: Intro interpretability research

TIMELINE · prerequisites → this → what builds on it

2017 Transformer &…2020 Logit Lens & Prob…2022 Circuits & In…2022 Locating & Ed…2023 Activation St…

01Original paper

interpreting GPT: the logit lens

nostalgebraist · LessWrong · 2020

Open ↗ ✓ Verified · 官方页面 HTTP/WebFetch 核实

02Where it sits

Learn the left column first; the right column builds on this one.

Learn first

Transformer & Self-Attention

Logit Lens & Probing

Then you can learn

Left: prerequisites · right: builds on this · dashed: related. Full map →

03Learning resources

Official docs, explainers, courses and reference code. All links checked.

Explainers

nostalgebraist：interpreting GPT: the logit lens ↗EN

Reference code

TransformerLens ↗EN

04Where the course teaches it

U91 Interpretability: from Attribution to Circuits Attribution, probes, the logit lens, superposition, SAEs, circuit tracing, knowledge editing and activation steering. 10 h9 algorithms4 paperslesson ready

For AI / data

Verification: verified, checked 2026-10-21.
JSON: /api/learn/algorithms.json (id: logit-lens)