“GenAI systems don’t give attribution to source materials because at least as constituted now, they can’t.”
Our reading:
Current generative AI systems do not attribute source materials because, as currently built (black boxes), they cannot.
Assessment: Unresolved
What happened
Does the evidence support the cause?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
A June 3, 2023 paper demonstrated training-data influence analysis using specially trained diffusion-model ensembles [e1]. GitHub's November 30, 2023 update described a Copilot public beta that provided references to matching public code [e2]. These demonstrations do not establish reliable training-source attribution in generative AI systems deployed by December 29, 2023 [e1, e2]. Neither establishes that technical inability explains missing attribution, so the causal claim remains unresolved [e1, e2].
Caveats
- The research demonstration does not establish reliable attribution in contemporary deployed systems [e1].
- Code matching identifies similar material without proving which training sources influenced an output [e2].
- Neither item establishes why training-source attribution was absent across the claimed systems [e1, e2].
Evidence (2 items)
Context only: arXiv; Zheng Dai and David K Gifford,
assessing influence.
Why it matters: The researchers proposed specially trained ensembles that permit analysis of training-data influence. This limits a sweeping impossibility argument, but does not demonstrate reliable attribution in the deployed DALL-E or ChatGPT systems.
- Publisher
- arXiv; Zheng Dai and David K Gifford
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
Context only: GitHub,
matches a Copilot suggestion.
Why it matters: A contemporary generative product could identify matching public material and expose references. Matching establishes similarity, however, rather than proving which training examples caused an output.
- Publisher
- GitHub
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Causal claim
- Deadline
- None given. We don’t add one.
Conditions
- as currently constituted (their words)
Where we had to interpret
- “Now” refers to systems available on 2023-12-29.
- In this discussion of training materials and provenance, attribution means reliably identifying the training sources behind an output. Merely citing a retrieved webpage or labeling an image AI-generated does not establish that capability.
- The claim asserts that technical inability explains missing attribution. Missing product features alone cannot establish that cause.
- A limited research demonstration or matching service bears against an absolute impossibility claim, but does not necessarily establish reliable attribution in deployed systems.
Right if: Technical evidence establishes that contemporary deployed systems could not reliably identify the training sources behind their outputs, and that this limitation explains their failure to provide attribution.
Wrong if: A contemporary deployed system demonstrably provides reliable training-source attribution, or technical evidence establishes that the claimed inability was not the reason attribution was absent.
We wrote this test after the outcome was known.
What would settle it
Technical evidence addressing systems available on December 29, 2023 that establishes whether inability to identify training sources reliably caused missing attribution; or a validated contemporary deployed system demonstrating reliable training-source attribution.
We found 16 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Things are about to get a lot worse for Generative AI, paragraph 26.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-29. The post may have been edited after 2023-12-29.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Things are about to get a lot worse for Generative AI
GenAI systems don’t give attribution to source materials because at least as constituted now, they can’t.
- Where
- paragraph 26
- Published
- 2023-12-29
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: our summary didn’t match their words.