Resolution criteria
Resolves to the number of distinct confirmed errors in the video, posted in the comments before close (January 1, 2027).
An error is a year, caption, quote or figure shown in the video that contradicts its primary source. As each entry is sung, a stamp shows its year and caption with its source's domain underneath, and the counter at top left shows the year (from 2025, with the quarter or month). Figures drawn on screen have their own sources, in the ledger's second table.
How I'll judge:
Primary sources win: the organization's own release, the paper, the filing, the official results. Wikipedia and the press don't count against them.
Any of a thing's dates will do. Where something has several (announced, preprinted, published, premiered, released, shut down), the stamp and the counter may show any of them. A date is an error only if I can't produce a primary source that gives it for that thing.
An undated number must hold for the stamp's year; a dated one ("games by 1993"), for its date. Where official sources disagree, either will do. Quotes must match word for word. Code and mock screens are illustrations, not claims.
Only the screen is judged. Lyric phrasing, sung respellings and omissions aren't errors. Slips in the ledger's fact column earn Ṁ200 but don't count as errors.
First to post a qualifying error gets it. My own finds count too, with no bounty. I'll do my best to rule within a day.
I made the video, I'm the judge and I'm paying, so I won't argue a loss: if your source beats mine, you win. I won't trade on this market.
Background
"We Didn't Write the Weights" is a four-minute music video chronicling 76 years of artificial intelligence history (1950 to 2026). The video stamps each of its 199 entries with a year, a caption and a source as it's sung, and draws most of them. The ledger behind it is public: We Didn't Write the Weights - The Ledger (every entry with its year, fact and source, and every figure drawn on screen with the sentence it comes from).
Running count: 0 confirmed errors.
On top of the market, I'll pay by managram:
Ṁ1,000 per confirmed error. Up to Ṁ5,000 total.
Ṁ200 for a primary source the ledger doesn't have that confirms an on-screen figure, a typo on screen, or a wrong fact in the ledger's fact column. One per entry, up to Ṁ2,000 total.
If you find an error, buy the matching bucket before you post it.
This is round two. The first market turned up two errors from readers and more from my own audit, all fixed in this cut.
1,000People are also trading
@bobalobascrob or is this because the old market closed early? is this additional errors? or just errors?
@bobalobascrob Not a hallucination (and understandable!), it's explained in the footnote at the bottom: this is round two for a new cut with corrections motivated by the prior closing early.
Five more primary-source additions for consideration under the M200-per-entry category. These confirm figures listed in the current ledger’s screen-figures table; I’m not claiming additional video errors.
1. Entry 38 (0:41), ICOT, 1982, 10-year initiative. ICOT’s own Kawanobe paper at FGCS 1984 confirms an April 1982 start and a ten-year plan in its abstract (PDF page 1): https://www.airc.aist.go.jp/aitec-icot/ICOT/Museum/FGCS/FGCS84en-prs/84ePRS2.pdf . The ledger currently cites Wikipedia.
2. Entry 64 (1:19), 2045. Ray Kurzweil’s own “So what is the singularity?” dialogue, identified on the page as written in 2005, explicitly gives 2045 in answer 3: https://www.writingsbyraykurzweil.com/so-what-is-the-singularity . This adds the author’s account to Wikipedia.
3. Entry 111 (2:11), about 6,000 compounds. The original Stokes et al. Cell paper specifies 6,111 molecules in the Drug Repurposing Hub, p.690 (PDF page 4, right column): https://people.csail.mit.edu/tommi/papers/Stokes_etal_Cell2020.pdf . This adds the research paper alongside the MIT News story already listed.
4. Entry 122 (2:20), ECCN 3A090. The actual BIS rule, 87 FR 62186, document 2022-21658, confirms the classification in section III.A and amendment 26; its footer also records filing on 7 October 2022: https://www.govinfo.gov/content/pkg/FR-2022-10-13/html/2022-21658.htm . The ledger already has the BIS announcement and a WSGR explanation; this adds the original regulatory text.
5. Entry 180 (3:27), Eurisko’s 96 ships and wins in 1981/1982. Lenat’s own 1983 EURISKO paper confirms the tournament years/results on pp.73–74 and the 96-ship fleet on p.80: https://www.cs.northwestern.edu/~mek802/papers/not-mine/Lenat_EURISKO.pdf . This adds the original research report to Wikipedia.
I checked all three current gist files: these underlying sources are additional, rather than alternate links to papers already listed.
Five primary-source additions for the M200 category, one per entry. These confirm figures listed in the current ledger’s screen-figures table; I’m not claiming video errors or additions to the error count.
1. Entry 35 (0:38), BKG 9.8, 7–1. Hans Berliner’s own 1980 paper confirms the result against Luigi Villa on 15 July 1979. The points score is 7–1; five games were played. Original paper text: https://bkgm.com/articles/Berliner/BackgammonProgramBeatsWorldChamp/
2. Entry 63 (1:18), Stanley, FINISHED 6:53:58. The Stanford/Volkswagen team’s AAAI 2006 paper gives exactly 6 hours 53 minutes 58 seconds in “Results from the Race,” printed p.986 (PDF page 5): https://cdn.aaai.org/AAAI/2006/AAAI06-154.pdf . This adds the original technical paper alongside the existing ScienceDaily/Stanford report.
3. Entry 110 (2:11), AlphaFold 2, 92.4 median GDT. DeepMind’s 30 November 2020 announcement, under “Results from the CASP14 assessment,” confirms a median 92.4 GDT across all targets: https://deepmind.google/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/
4. Entry 133 (2:30), WGA 148 days / SAG-AFTRA 118 days. WGA’s 14 March 2024 release explicitly describes the 2023 148-day strike: https://awards.wga.org/awards/awards-recipients/honorary-service-awards/morgan-cox-award/chris-keyser-david-goodman . SAG-AFTRA’s president and chief negotiator confirm 118 days in their 9 November 2023 letter: https://www.sagaftra.org/message-your-sag-aftra-president-and-chief-negotiator . Both durations are one entry.
5. Entry 134 (2:30), Bletchley Declaration, 28 countries. The UK host government’s 1 November 2023 release directly confirms that figure: Countries agree to safe and responsible development of frontier AI in landmark Bletchley Declaration . This specific release is additional to the general summit page already linked.
I checked these specific sources against both ledger tables; none of these URLs is currently listed.