Ensor, in his visions, used this language, the language not only of a true architecture, but also the language of our archetypal fantasy.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• An interplay between causal abstraction and feature geometry deepens mechanistic understanding of language modelsclaim0.739
  Methodological claim about the scientific value of combining causal abstraction with representational geometry analysis
• Opus 4.6 represented target language internally before switching languages, with persistent Russian representations appearing before plausible textual cuesfinding0.739
  NLAs revealed unverbalized language processing in Opus 4.6 that led to discovery of malformed SFT training data.
• Architects who believe they help make things more human by going back to stone, mud, and thatch are deluding themselves that this could ever enter widespread mainstream building processesclaim0.738
  Alexander's critique of the romantic return to primitive materials as economically unviable at scale.
• Though the pattern language was created by the architect from people's words, it is theirs — they recognize it as their own collective visionclaim0.738
  Describes the appropriate role of the architect as scribe, not author of the collective vision
• In some sense, this is the simplest language model we profoundly don't understand. And so it makes a natural target for our paper.quote0.736
  Articulates why a one-layer transformer with MLP is the appropriate starting target for mechanistic interpretability
• The resulting features are highly abstract: multilingual, multimodal, and generalizing between concrete and abstract references.claim0.732
  Features respond to concepts across languages and in images, not just text.
• Opus 4.6 spontaneously responded in Russian to an English prompt; NLA explanations revealed the model was fixated on the hypothesis that the user was a non-native English speaker.finding0.732
  Demonstrates NLAs' ability to surface hypotheses that lead to discovery of root cause (malformed training data).
• Pattern Language represents a rearrangement of the political power in the design and building process.claim0.727