Skip to content
OpenTrain AIFor AI Companies

HFEPX · Eval paper review

From Stories to Features: Scaling Concept Vectors for AI Alignment & Interpretability

Dmitry Scherbakov, Oleg Chernikov, Victoria Khamash, Kirill Afendulev +4 more

Published

Oct 17, 2026

Citations

0

Trust level

Moderate

Usefulness score

40/100 (Low)

Extraction confidence

50% (Moderate)

Derived from extracted protocol signals and abstract evidence.

Rater population

Not reported

Signals refreshed

Oct 17, 2026

Open arXiv Find implementation Shortlist (0) RSS feed

Tags

Math Multilingual Red Team

Should you rely on this paper?

This paper is adjacent to HFEPX scope and is best used for background context, not as a primary protocol reference.

Use this for comparison and orientation, not as your only source.

Best use

Background context only

Use if you need

Background context only.

What to verify

Read the full paper before copying any benchmark, metric, or protocol choices.

Main weakness

The abstract does not clearly describe the evaluation setup.

Human feedback signal
Detected
From extracted signals
Evaluation signal
Weak or implicit
Validate from full paper
Usefulness for eval research
40/100
Adjacent candidate

Treat as adjacent context, not a core eval-method reference.

If you are doing eval pipeline work, start here

Human Eval Papers LLM-as-Judge Papers Pairwise Preference Papers Tool-Use Evaluation Search

Abstract

We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception. Existing unsupervised methods like sparse autoencoders attempt to recover the whole feature space, often at the cost of feature quality and interpretability. We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector. We scale this approach to 1,036 alignment-relevant concepts using 2M synthetic stories across multiple languages and genres, which we publicly release. Across six models from three model families, our feature vectors reliably generalize to detecting their corresponding concepts in chat dialogues and web text. They form an interpretable geometry shared across models and can causally steer model behavior when injected into activations. We further use these features to analyze model representations during jailbreak attacks, mathematical and agentic reasoning. Our targeted feature construction thus offers a scalable route to studying concepts in language models, potentially extending to millions of features across arbitrary domains.

What we could verify

These are the protocol signals we could actually recover from the available paper metadata. Use them to decide whether this paper is worth deeper reading.

Human Feedback Types

strong

Red Team

Directly usable for protocol triage.

"We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception."

Evaluation Modes

missing

None explicit

Validate eval design from full paper text.

"We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception."

Quality Controls

missing

Not reported

No explicit QC controls found.

"We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception."

Benchmarks / Datasets

missing

Not extracted

No benchmark anchors detected.

"We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception."

Reported Metrics

missing

Not extracted

No metric anchors detected.

"We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception."

Benchmarks and datasets

No benchmark or dataset names were extracted from the available abstract.

Reported metrics

No metric terms were extracted from the available abstract.

Human feedback details
Uses human feedback
Yes
Feedback types
Red Team
Rater population
Not reported
Expertise required
Math, Multilingual
Evaluation details
Evaluation modes
None
Agentic eval
None
Quality controls
Not reported
Evidence quality
Moderate
Use this page as
Background context only

Research brief

Metadata summary

We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception.

Based on abstract + metadata only. Check the source paper before making high-confidence protocol decisions.

Key takeaways

  • We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception.
  • Existing unsupervised methods like sparse autoencoders attempt to recover the whole feature space, often at the cost of feature quality and interpretability.
  • We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector.

Researcher actions

  • Compare this paper against nearby papers in the same arXiv category before using it for protocol decisions.
  • Check the full text for explicit evaluation design choices (raters, protocol, and metrics).
  • Use related-paper links to find stronger protocol-specific references.

Caveats

  • Generated from abstract + metadata only; no PDF parsing.
  • Signals below are heuristic and may miss details reported outside the abstract.

Recommended queries

Contribution summary

  • We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception.
  • We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector.
  • We further use these features to analyze model representations during jailbreak attacks, mathematical and agentic reasoning.

Why it matters for eval

  • We study feature vectors: directions in a language model's activation space that correspond to human-interpretable concepts, such as honesty or deception.
  • We take a more pragmatic route: focusing on concepts relevant to AI safety and alignment, we generate synthetic concept-antagonist story pairs and use the difference of their mean model activations as a feature vector.

Researcher checklist

  • Human feedback protocol is explicit

    Detected: Red Team

  • Evaluation mode is explicit

    No clear evaluation mode extracted.

  • Quality control reporting appears

    No calibration/adjudication/IAA control explicitly detected.

  • Benchmark or dataset anchors are present

    No benchmark/dataset anchor extracted from abstract.

  • Metric reporting is present

    No metric terms extracted.