500'000€ Prize for Compressing Human Knowledge
500'000€ Prize for Compressing Human Knowledge
(widely known as the Hutter Prize.<br>Total payout so far: 29'945€)
(widely known as the Hutter Prize) --><br>-->
Compress the 1GB file enwik9 to less than the current record of about 110MB
Introduction<br>--><br>The Task
Motivation
Detailed Rules for Participation
Previous Records
More Information
Discussion forum on the contest and prize
Tips<br>--><br>History
Committee
FAQ<br>--><br>Donations<br>--><br>Frequently Asked Questions
Contestants<br>Links
Disclaimer
News: The H-Prize went BIG!<br>Why?
10×Size, 10×Purse, 10×Time, 10×RAM, 10×HDD.
Who will get here first?
--><br>News: Kaido Orav & Byron Knoll are
the eighth Winners! Congratulations!
... the contest continues ...
This site is under construction.
The prize money is not officially available yet
(I first have to sell my Ferrari :-)<br>-->
Introduction<br>-->
Being able to compress well is closely<br>related to intelligence as explained below.<br>While intelligence is a slippery concept,<br>file sizes are hard numbers.<br>Wikipedia is an extensive snapshot of Human Knowledge.<br>If you can compress the first 1GB of Wikipedia better<br>than your predecessors, your (de)compressor likely has to be smart(er).<br>The intention of this prize is to encourage development<br>of intelligent compressors/programs as a path to AGI.
Interview with Lex Fridman (26.Feb'20)<br>(Video,<br>Audio,<br>Tweet)
The Task
Losslessly compress the 1GB file enwik9 to less than 110MB. More precisely:<br>enwik9 of less than 116MB. More precisely: -->
Create a Linux or Windows compressor comp.exe of size S1 that compresses enwik9 to archive.exe of size S2<br>such that S:=S1+S2 L := 110'793'128 (previous record)5'000×(days passed since 1.1.2021) -->.<br>If run, archive.exe produces (without input from other sources) a 109 byte file<br>that is identical to enwik9.<br>If we can verify your claim,<br>you are eligible for a prize of 500'000€×(1-S/L). Minimum claim is 5'000€ (1% improvement).<br>Restrictions: Must run in ≲50 hours using a single CPU core and test machine.
Remark: You can download the zipped version<br>enwik9.zip of enwik9<br>here (≈300MB).<br>Please find more details including constraints and relaxations at<br>http://prize.hutter1.net/hrules.htm.
Motivation
This compression contest is motivated by the fact that<br>being able to compress well is closely<br>related to acting intelligently, thus reducing<br>the slippery concept of intelligence to hard file size numbers.<br>In order to compress data, one has to find regularities in them,<br>which is intrinsically difficult (many researchers live from<br>analyzing data and finding compact models).<br>So compressors beating the current "dumb" compressors need to be<br>smart(er). Since the prize<br>wants to stimulate developing "universally" smart compressors,<br>we need a "universal" corpus of data.<br>Arguably the online encyclopedia<br>Wikipedia<br>is a good snapshot of the Human World Knowledge.<br>So the ultimate compressor of it should "understand"<br>all human knowledge, i.e. be really smart.<br>enwik9 is a hopefully representative 1GB extract from Wikipedia.
Detailed Rules for Participation
Rules<br>Relaxations<br>Participation<br>Award<br>More Information
Previous Records<br>--><br>Baseline Enwik9 and Previous Records Enwik8
Author (enwik9)<br>Date<br>Decompressor<br>Total Size<br>Compr.Factor|RAM|time<br>% | Award<br>Sponsor
You?<br>202?
>9.12 |<br>>1%| >5'000€<br>Marcus Hutter
Kaido Orav & Byron Knoll<br>3.Sep 2024<br>fx2-cmix<br>110'793'128<br>9.03 | 95% | 99%<br>1.59%| 7'950€<br>Marcus Hutter
Kaido Orav<br>2.Feb 2024<br>fx-cmix<br>112'578'322<br>8.88 | 8.9GB | ~50h<br>1.38%| 6'911€<br>Marcus Hutter
Saurabh Kumar<br>16.Jul 2023<br>fast cmix<br>114'156'155<br>8.76 | 8.4GB | 43h<br>1.04%| 5'187€<br>Marcus Hutter
Artemiy Margaritov<br>31.May 2021<br>starlit ...<br>115'352'938<br>8.67 | 10GB | ~50h<br>1.1% | 9000€<br>Marcus Hutter
Alexander Rhatushnyak<br>4.Jul 2019<br>phda9v1.8 ...<br>116'673'681<br>8.58 | 6.3GB | ~23h<br>-- | pre-prize
Author (enwik8)<br>Date<br>Decompressor<br>Total Size<br>Compr.Factor|RAM|time<br>% | Award<br>Sponsor
You?
-->
Alexander Rhatushnyak<br>4.Nov 2017<br>phda9 ...<br>15'284'944<br>6.54 | 1048MB | ~5h<br>4.17%| 2085€<br>Marcus Hutter
Alexander Rhatushnyak<br>23.May 2009<br>decomp8 ...<br>15'949'688<br>6.27 | 936MB | ~9h<br>3.2%| 1614€<br>Marcus Hutter
Alexander Rhatushnyak<br>14.May 2007<br>paq8hp12 -7<br>16'481'655<br>6.07 | 936MB | 9h<br>3.5%| 1732€<br>Marcus Hutter
Alexander Rhatushnyak<br>25.Sep.2006<br>paq8hp5 -7<br>17'073'018<br>5.86 | 900MB | 5h<br>6.8%| 3416€<br>Marcus Hutter
Matt Mahoney<br>24.Mar.2006<br>paq8f -7<br>18'324'887<br>5.46 | 854MB | 5h<br>-- | pre-prize
More Information
Discussion forum on the contest and prize<br>Compression benchmarks enwik9 and others<br>Motivation of compressing the Human Knowledge<br>Detailed rules for participation --><br>Information about the enwik9 data file<br>Wikipedia on the Hutter Prize
Tips<br>-->
History
13??:<br>William of Ockham's razor: Entities should not be multiplied beyond necessity.<br>1964:<br>Ray Solomonoff introduced algorithmic probability for universal prediction.<br>1996:<br>Leonid Broukhis...