“Still, the usual reports of silly errors and embarrassing results trickled out within hours.”
Our reading:
The usual reports of silly errors and embarrassing results from Gemini 3 trickled out within hours of Google announcing it.
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
Two firsthand reports document obvious errors: incorrect transcription timestamps and rejection of genuine information about 2025. A Techmeme snapshot places both reports in public circulation by 5 p.m. Eastern on November 18, six hours after the 11 a.m. launch coverage. This establishes an interval measured in hours, rather than relying on calendar dates. Karpathy's version caveat limits conclusions about the released model, but Marcus claimed that reports appeared.
Caveats
- The launch-time anchor is contemporary news coverage, not a company timestamp specifying the exact announcement minute.
- The archive supplies an upper bound on circulation, not exact publication times.
- Karpathy suspected an earlier model version with a stale system prompt.
- These examples establish reports of errors, not their frequency or Gemini 3's overall reliability.
Evidence (4 items)
Supports the claim: Simon Willison,
the timestamps do not line up
Why it matters: Willison identifies Gemini 3 Pro and shows a transcript placing adjournment near one hour when it occurred after three and a half hours. Item e4 establishes that his report was circulating within hours of launch.
- Publisher
- Simon Willison
- Kind of source
- Demonstration
- About events on
- Published
Supports the claim: Andrej Karpathy,
it was all fake.
Why it matters: Karpathy publicly reported Gemini 3 rejecting genuine information about 2025. He suspected an earlier version with a stale system prompt; the relevant event date is publication of his report, whose timing is independently bounded by e4.
- Publisher
- Andrej Karpathy
- Kind of source
- Demonstration
- About events on
- Published
Supports the claim: Techmeme,
5:00 PM ET, November 18, 2025.
Why it matters: The snapshot contains Karpathy's account of rejecting 2025 and Willison's announcement of his transcription test. Its explicit Eastern-time archive timestamp places both reports in circulation by six hours after the launch coverage in e1.
- Publisher
- Techmeme
- Kind of source
- Secondary summary
- Math
- 17:00 − 11:00 = 6 hours (inputs: e1: November 18, 2025, 11:00 a.m. EST launch coverage; e4: November 18, 2025, 5:00 p.m. ET archive containing both reports; No rounding; this is an archive-based upper bound, not either report's exact publication delay.)
- About events on
- Published
Context only: Semafor,
The Gemini 3 launch
Why it matters: The article reports the launch at 11:00 a.m. EST. This contemporary news timestamp supplies the announcement-time anchor; the reviewed company announcement gives only the date.
- Publisher
- Semafor
- Kind of source
- News report
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- the usual reports (their words)
Where we had to interpret
- The claim concerns when reports became public, not when the reported tests were performed.
- The ordinary reading is that multiple concrete reports of obvious mistakes appeared a short number of hours after the announcement. Calendar dates alone are insufficient.
- Reports from early-access testing can count, but uncertainty about the tested version must be disclosed.
- The target period covers the November 18 announcement and the November 19 post. Evidence within that period must additionally satisfy the elapsed-hours requirement.
Right if: Concrete public reports of plainly mistaken Gemini 3 outputs, with timestamps or contemporary archives establishing their appearance within a short interval measured in hours after the announcement.
Wrong if: Reliable records establish that the purported early reports concerned other models, were fabricated, or appeared substantially later, and sufficiently complete contemporaneous records rule out qualifying early reports.
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, Hot take on Google’s Gemini 3, paragraph 1.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2025-11-19.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Hot take on Google’s Gemini 3
Still, the usual reports of silly errors and embarrassing results trickled out within hours.
- Where
- paragraph 1
- Published
- 2025-11-19
- 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.