“Fundamentally, Southen and I had argued that our images were evidence not only that some popular GenAI systems had trained on copyrighted images (nobody is really disputing this)”
Our reading:
Some popular GenAI systems have trained on copyrighted images.
Assessment: Supported
What happened
Was it true for the period described?
Outcome: Supported
The evidence matches what they said.
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
The court record confirms that copyrighted works were used to train Stable Diffusion versions released before the post. A contemporaneous company announcement documents widespread adoption. This supports the limited claim that some popular image generators trained on copyrighted images. It does not establish that the training infringed copyright.
Caveats
- The clearest training admission was published after the post, but concerns earlier training and was available before the evaluation date.
- The adoption figure is company-reported.
- This evidence does not establish the training contents of every generator or independently validate Marcus and Southen's image experiments.
Evidence (2 items)
Supports the claim: High Court of Justice of England and Wales,
“many” Copyright Works
Why it matters: The judgment records admitted training on copyrighted works, including training of v2.0, which paragraph 39 dates to a November 2022 release. The event date refers to that earlier training, not the later admission or judgment.
- Publisher
- High Court of Justice of England and Wales
- Kind of source
- Regulatory filing
- About events on
- Published
Supports the claim: Stability AI,
more than 200,000 developers globally.
Why it matters: The company's reported adoption establishes widespread use before Marcus's post. Together with e1, it supplies a documented example meeting both parts of the claim.
- Publisher
- Stability AI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The claim concerns training that had occurred by publication on 2024-01-24. It specifies no starting date.
- “Some” is an existence claim: a documented example of a popular image generator is sufficient.
- “Popular” means widely used. No particular user count or market share is required.
- Copyrighted images include licensed copyrighted images. The claim does not require proof of unauthorized use or copyright infringement.
- This assessment tests the stated training fact, rather than whether Marcus and Southen's images alone established it.
Right if: Reliable records establish that at least one widely used generative image system had trained on images protected by copyright by 2024-01-24.
Wrong if: Comprehensive reliable records establish that no popular generative image system had trained on copyrighted images by 2024-01-24. Permission to use copyrighted images would not contradict the claim.
We wrote this test after the outcome was known.
We found 10 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, No, multimodal ChatGPT is not going to “trivially” solve Generative AI's copyright problems, paragraph 4.
- 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 2024-01-24.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, No, multimodal ChatGPT is not going to “trivially” solve Generative AI's copyright problems
Fundamentally, Southen and I had argued that our images were evidence not only that some popular GenAI systems had trained on copyrighted images (nobody is really disputing this)
- Where
- paragraph 4
- Published
- 2024-01-24
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Supported. 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
A reviewer said our test wasn’t fair to what was actually said.