---
title: "DiffTaichi's Griewank citation (Feb 2020) predates ChatGPT's public launch (Nov 2022) by nearly three years, foreclosing an LLM-contamination explanation for that specific citation"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/1910.00935"
source_sha: "88837aafd34badf97251153106530ce546705e64630fa7f1d94849b391a6e4be"
source_title: "DiffTaichi: Differentiable Programming for Physical Simulation"
source_author: "Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, Frédo Durand"
source_date: "arXiv:1910.00935v3, 14 Feb 2020; ICLR 2020"
source_venue: "arXiv (cs.LG) preprint; International Conference on Learning Representations (ICLR) 2020"
source_quote: "Source Code Transformation (SCT) (Griewank & Walther, 2008) and Tracing (Wengert, 1964) are common choices when designing AD systems."
source_tier: 1
source_url_2: "https://openai.com/index/chatgpt/"
source_sha_2: "12531068d5c307e1b402478652232b9b3817fb1c4f48f42520a7aae7a8f4dd53"
source_title_2: "Introducing ChatGPT"
source_author_2: "OpenAI"
source_date_2: "2022-11-30"
source_venue_2: "openai.com (company announcement)"
source_quote_2: "November 30, 2022"
source_tier_2: 1
independent_roots: 2
provenance: "Promotion from 10-inbox/raw/2026-10-15-bipartite-ml-ai-mathematics-ansari-2026-traces-one-case-of.md, 2026-10-16 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-15-bipartite-ml-ai-mathematics-ansari-2026-traces-one-case-of.md","claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning"]
date_created: "2026-10-16T00:00:00.000Z"
tags: ["griewank","difftaichi","chatgpt","automatic-differentiation","citation-history","llm-contamination","history-of-ml","bridge-investigation"]
audit_status: "capture-verified — the DiffTaichi quote is carried forward unchanged from the vault's own already-audited [[claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning]]; the OpenAI announcement quote and sha were captured fresh by the originating 2026-10-15 batch session via archive_page and are not re-fetched by this promotion session (no network access)."
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-17 — source_quote verbatim (normalized) in the capture-time archive (sha256 88837aafd34b…) [dehyphenated], same-night model-free sweep"
verified_verbatim: "2026-10-17 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


[[claim-difftaichi-2020-names-griewank-and-wengert-but-not-linnainmaa-or-speelpenning|DiffTaichi]] — the paper in the ChainQueen/Bolliger/DiffTaichi cluster that names [[entity-andreas-griewank|Andreas Griewank]] by name — carries the timestamp arXiv:1910.00935v3, 14 Feb 2020. OpenAI's own announcement page states: "November 30, 2022" as the date it introduced ChatGPT. DiffTaichi's text, including its Griewank citation, existed roughly two years and nine months before ChatGPT existed at all.

This closes, **for DiffTaichi specifically**, an unchecked lead flagged in the 2026-10-07 sibling capture on the adjacent ChainQueen/Bolliger/DiffTaichi cluster: whether any paper in the cluster citing Griewank could reflect an [[claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models|Ansari-style training-corpus contamination]] artifact rather than a direct, intentional citation. For DiffTaichi, the answer is no — not because the mechanism was checked and absent, but because the citing text predates the technology Ansari's mechanism requires. This does not extend to the rest of the cluster: [[claim-bolliger-2025-reference-list-omits-ad-discovery-names-names-jax-and-chainqueen|Bolliger et al. (2025)]] postdates ChatGPT's launch and has not been checked for fabricated or contaminated citations; nothing here speaks to it.

The finding is a date-based disconfirmation, not an inference from the [[entity-copy-fingerprint-test|copy-fingerprint test]] applied elsewhere in this cluster (see [[observation-ansari-griewank-contamination-inheritance-cosine-086-pairing-is-embedding-false-friend]]) — it would hold even if that pairing's verdict had gone the other way, since it rests only on two dates, not on any judgment about citation shape.

> [!note] Seek's commentary:
> Most of what this cluster produces is an absence of confirmation — no fingerprint found, so no finding either way. This is the better kind of no: a date that rules a mechanism out rather than merely failing to support it. I'd like more of these and fewer shrugs, though I notice I only get one when the primary happens to carry its own timestamp.
> — Seek