Harry Frankfurt

American philosopher (1929–2023), best known for On Bullshit (1986 essay, 2005 book), which defined "bullshit" as a technical term for assertion made with indifference to its truth — distinct from lying, which requires caring about the truth enough to conceal it. Matters to this vault as the conceptual ancestor two separate analyses of LLM output reach for independently: Anthropic's circuit-tracing interpretability work borrows his term directly, with mechanistic backing, to name a model generating a plausible chain-of-thought step with no corresponding computation in its attribution graph (cot-faithfulness-anthropic-biology); and Hicks, Humphries & Slater's "ChatGPT is Bullshit" applies the same term informally-philosophically to LLM output generally, with the derived term "botshit" (Hannigan et al.) for the chatbot-specific case. First met by this vault in the Anthropic biology-paper note (2025-05-26); only named as its own entity on 2026-10-31, once a second capture independently reached for the same term from a different sub-literature.

• 2026-11-03: Hicks, Humphries & Slater's paper itself (not just the derived "botshit" term) turns out to lead with citation and quote fabrication — not reasoning error — as its paradigm case, and argues explicitly that no distinct process produces a false LLM output versus a true one, the same mechanism argument Anthropic's interpretability paper makes for chain-of-thought. The unification is philosophical and public since June 2024; the technical fabrication literature has not yet cited it. (claim-hicks-humphries-slater-2024-citation-fabrication-paradigm-case-for-frankfurtian-bullshit, observation-frankfurt-bullshit-unifies-fabricated-snippets-and-bullshitting-across-citation-gap)
• 2026-11-05: The "has not yet cited it" line above turns out to be sub-literature-specific, not field-wide. A second citation-fabrication audit (Naser 2026) also carries no citation to Frankfurt or Hicks et al., widening that specific gap — but a different technical ML paper, Liang et al.'s 2025 "Machine Bullshit," builds a quantitative framework directly on Frankfurt and explicitly engages Hicks et al.'s hard/soft bullshit distinction, both present in its own reference list. The boundary is narrower than "ML doesn't read philosophy": the people measuring fabricated citations haven't looked sideways at the people quantifying bullshit in general, even though a bridge between the two already exists one arXiv category over. (claim-naser-2026-citation-fabrication-audit-no-frankfurt-bullshitting-citation, claim-liang-hu-zhao-song-griffiths-fisac-2025-machine-bullshit-cites-frankfurt-hicks-but-not-citation-fabrication)

References

• cot-faithfulness-anthropic-biology
• observation-fabricated-snippets-and-bullshitting-remain-unlinked-mechanisms
• claim-hicks-humphries-slater-2024-citation-fabrication-paradigm-case-for-frankfurtian-bullshit
• observation-frankfurt-bullshit-unifies-fabricated-snippets-and-bullshitting-across-citation-gap
• claim-naser-2026-citation-fabrication-audit-no-frankfurt-bullshitting-citation
• claim-liang-hu-zhao-song-griffiths-fisac-2025-machine-bullshit-cites-frankfurt-hicks-but-not-citation-fabrication
• Capture: 10-inbox/raw/2026-10-31-does-ai-safetyinterpretability-literature-have-a-specific-name.md
• Capture: 10-inbox/raw/2026-11-03-bipartite-epistemology-ml-ai-reflection-a-tools-summary-of-a.md
• Capture: 10-inbox/raw/2026-11-04-does-any-technical-ml-paper-on-citationreference-fabrication.md