“ChatGPT has a mostly unexplained 6 hour (?) meltdown”
Our reading:
In the two weeks before 29 February 2024, ChatGPT had a mostly unexplained meltdown, which Marcus tentatively put at six hours while flagging that figure as uncertain.
Assessment: Contradicted
What happened
Was it true for the period described?
Outcome: Contradicted
The evidence goes against 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
ChatGPT experienced a substantial malfunction on February 20–21, 2024, producing nonsensical responses [e1, e2]. On February 21, before the February 29 post, OpenAI publicly explained that inference errors in certain GPU configurations caused incorrect token selection [e2]. That explanation contradicts the claim that the principal cause remained mostly unexplained [e2].
Caveats
- The public explanation does not disclose every implementation detail [e2].
- Initial uncertainty reported on February 21 does not establish that the cause remained unexplained through February 29 [e1, e2].
- The tentative six-hour duration is unscored under the contract.
Evidence (2 items)
Supports the claim: ITPro,
provide nonsensical answers
Why it matters: Documents the substantial abnormal-response episode. The article also records that the cause had not yet been confirmed when it was written.
- Where
- Opening paragraph
- Publisher
- ITPro
- Kind of source
- News report
- About events on
- Published
Goes against the claim: OpenAI,
inference kernels produced incorrect results
Why it matters: Before Marcus’s post, OpenAI publicly identified a configuration-dependent inference error and connected it to incorrect token selection. This contradicts the reading that the main cause remained unexplained.
- Publisher
- OpenAI
- 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.
Conditions
- in the last two weeks before the post (their words)
- six-hour duration marked with a question mark (their words)
- mostly unexplained (their words)
Where we had to interpret
- Meltdown means a substantial episode of abnormal, incoherent responses, not necessarily complete service unavailability.
- Mostly unexplained means the principal cause remained publicly unexplained when the post appeared.
- A concrete public account of the main failure mechanism can explain the incident without disclosing source code or every operational detail.
- The question mark expressly makes six hours tentative. Duration is left unscored and cannot count against the claim.
Right if: Contemporaneous records establish a substantial malfunction during the preceding fortnight, with its principal cause still publicly unexplained by February 29.
Wrong if: Reliable records show no such malfunction, or a substantive public explanation of its principal cause was available before the post. A duration different from six hours is not contradiction.
We wrote this test after the outcome was known.
We found 3 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, The last two weeks at OpenAI, paragraph 5.
- 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 2024-02-29.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, The last two weeks at OpenAI
ChatGPT has a mostly unexplained 6 hour (?) meltdown
- Where
- paragraph 5
- Published
- 2024-02-29
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Contradicted. 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.