Sparse autoencoders are powerful tools for discovering previously unknown concepts represented in language models.
Structured and assessed
First pass. Read the position paper (arXiv 2506.23845 v2) sections 1-5 and its cited cases. Built two arguments: for (discovery case studies; new subclaim on rhyme planning, plus the automated-audit ablation) and against (matched comparisons: AuditBench null, model-diffing parity, embedding-search replication). Placed both unplaced passages on the 'for' argument (second deduplicated). Assessed inconclusive (0.6), no credence (evaluative, two readings pull apart). Importance 0.55, contestation 0.65. Relied on the sibling question's full readings of the audit studies rather than re-reading them. Did not read the Biology paper or the persona-feature work; marginal yield 0.4.
Assessed Inconclusive
verdict confidence 0.60
The claim is the thesis of a widely cited 2025 position paper (Peng, Movva, Kleinberg, Pierson and Garg, ICML 2026), which argues that sparse autoencoders (SAEs), though they lose to simple baselines when acting on concepts specified in advance, are powerful tools for discovering concepts no one specified. The evidence shows that SAEs and related sparse dictionaries can surface such concepts. It does not yet show that they are powerful at it in the sense the paper intends, a comparative advantage over other methods, and where that has been tested the results are mostly null. **The case for.** A sparse dictionary lists many interpretable features without being told what to look for, and a researcher can select those that meet a criterion and check them by other means. The paper's strongest example inside a language model is Anthropic's 2025 attribution-graph work, in which transcoder features showed a model choosing its rhyme word before writing the line, a finding then confirmed by interventions whose effects show in the model's output. For mechanisms at this level of detail no simpler method has produced comparable accounts, which is the best support for the claim. In a hidden-objective audit, adding SAE tools raised an automated agent's success from near zero to 10 to 20%. **What the cited cases show and do not show.** The paper reports no experiment of its own. Its main measured win, hypothesis generation, fits SAEs to text embeddings of a dataset and finds concepts that predict an outcome; that is discovery about a corpus, not about concepts a language model represents, and its baselines were built by overlapping authors. The rhyme-planning result used transcoders rather than SAEs (the paper groups them together), was not compared with any alternative, and confirms a broader idea, that models carry information about tokens yet to come, which earlier probing work had already shown. The model-diffing papers it cites compare model outputs, a task on which a later study found a strengthened language-model baseline performing about as well. The audit gain depended on the dictionary having been trained on the documents that described the hidden objective, and vanished without them. **Matched comparisons.** On AuditBench, 56 models each trained with an unstated hidden behaviour, an SAE tool gave no benefit while prompting-based elicitation tools raised success from about 36% to above 50%; that SAE tool was weak (off-the-shelf dictionaries, a static readout), so the null does not rule out a stronger setup. In the blind audit where an SAE team was fastest, its authors note that embedding search could have replicated the discovery, and a second SAE team was slower than the team without SAEs. **Where it stands.** Read as "SAEs can surface concepts researchers had not specified, worth testing as hypotheses," the claim is supported. Read as the paper means it, a distinctive power to discover what other methods miss, no matched study yet shows it outside one audit whose dictionary was trained on the answer. What would settle it: a discovery task, on properties that arose in ordinary training rather than being planted, in which a dictionary and tuned simpler methods are run by the same blind investigators and graded by the same rule.
Claim entered the graph