Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Herzliya

EchoState

Guy Nutman · Team EchoState

Submitted to Digital Minds Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: EchoState

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

EchoState is an evaluation harness for testing whether a language model's description of its own state has any relationship to that state. It builds concept directions contrastively — the difference between mean activations on opposing prompt sets — injects them into the residual stream mid-forward using PyTorch hooks, and measures three things: how far the internals moved, how far the output moved, and whether the model reported the induced state.

Every steered run is paired with a magnitude-matched random direction. Across six models spanning three architecture families (82M–500M parameters, 216 steered runs), concept directions elicited concept vocabulary in 45% of runs against 0% for random directions of identical magnitude — despite the random directions producing larger activation divergence. The effect is attributable to direction rather than perturbation strength, and is dose-dependent.

We argue this is a result about steering, not evidence of introspection: concept steering raises the probability of concept tokens directly, so an elicited report is confounded with the intervention's mechanical effect on the output distribution. The submission documents that confound, specifies what would constitute genuine evidence of introspection, and provides the tooling — a model-agnostic engine interface, contrastive direction construction, and matched controls — that makes the distinction testable.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. - Overall, I think this is a nice document: well structured, clear, and pleasant to read (no weird LLM sentences-thank you for that!).

    - The central idea is useful, but not especially novel. Prior work such as Lindsey already recognizes that making a model talk about an injected concept is not by itself evidence of introspection.

    - Still, I like that the paper turns this concern into explicit metrics and a reusable evaluation setup. This seems potentially useful for checking steering experiments more systematically.

    - I would have liked more motivation for the choice of positive affect, uncertainty, and self-reference. I am more familiar with interventions using more concrete concepts such as bread or love.

    - So I see this as a well-conducted and potentially useful (but minor) methodological contribution.

  2. The control experiment is well designed, and the self-critique is valid.

    However, you suggested three experiments but did not run any of them.

    Also, explain why the random group performs worse than the normal baseline.

    This unexpected result needs investigation and may be more important than the main finding.

Cite this project

@misc{nutman2026echostate,
  title = {{EchoState}},
  author = {Guy Nutman},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/echostate-ae7y}},
  url = {https://apartresearch.com/sprints/projects/echostate-ae7y}
}

More from Digital Minds Research Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026