Skip to content
Longterm Wiki

Mechanistic interpretability work

resource

Metadata

Source Tableresources
Source ID028435b427f72e06
DescriptionNeel Nanda's personal research homepage focused on mechanistic interpretability of neural networks, aiming to reverse-engineer how transformers and other models implement algorithms internally. His work includes foundational contributions like the discovery of grokking phenomena, superposition in ne…
Source URLwww.neelnanda.io/
Children
CreatedFeb 21, 2026, 3:56 PM
UpdatedMay 18, 2026, 12:24 PM
Synced

Record Data

id028435b427f72e06
urlwww.neelnanda.io/
titleMechanistic interpretability work
typeweb
summaryNeel Nanda's personal research homepage focused on mechanistic interpretability of neural networks, aiming to reverse-engineer how transformers and other models implement algorithms internally. His work includes foundational contributions like the discovery of grokking phenomena, superposition in ne
review
abstract
keyPoints
[
  "Leads mechanistic interpretability research at Google DeepMind, previously at Anthropic, focusing on understanding transformer internals",
  "Developed TransformerLens, a widely-used open-source library for interpretability research on GPT-style language models",
  "Contributed foundational wor…
publicationId
authors
authorEntityIds
publishedDate
tags
[
  "interpretability",
  "technical-safety",
  "ai-safety",
  "alignment",
  "evaluation",
  "capabilities"
]
localFilename
credibilityOverride
fetchedAt
contentHashf5a772ac97743d43
stableIdsid_VWiB2bX3Fb
fetchStatusok
lastFetchedAtMay 18, 2026, 12:24 PM
archiveUrl
stance
contextNoteNeel Nanda is one of the most prolific researchers in mechanistic interpretability; his homepage aggregates papers, blog posts, and tools that are frequently cited as entry points into the field.
resourcePurposehomepage
resourceSubtypehomepage
typeMetadata
publisherEntityId
relatedEntityIds
enrichmentStatusenriched
enrichmentDateMar 20, 2026, 11:14 PM
importanceScore0.72
contentLifecycle
Debug info

Thing ID: sid_VWiB2bX3Fb

Source Table: resources

Source ID: 028435b427f72e06