|
| TheRealPomax wrote:
| Q: Why do you restrict to a single CPU core and exclude GPUs?
| A: The primary intention is to limit compute and memory to some
| generally available amount in a transparent, easy, fair, and
| measurable way. 100 hours on one i7 core with 10GB RAM seems to
| get sufficiently close to this ideal
|
| Sorry, who are these people that don't have a GPU? Even laptops
| have GPUs. Why would you spend 100 hours on an "i7" (which
| generation? 4790K or six times faster 13700k?) CPU when you can
| achieve orders of magnitude better performance on a consumer GPU
| that literally everyone has access to?
| lucb1e wrote:
| Note that the competition close to 20 years old
|
| ...though I also had a GPU in 2006, so idk. Then again, you
| need to define _something_ as reference hardware and it doesn
| 't really matter what it is. Better compression should win out
| over less-good compression no matter if you run both on a
| 100-core system or a 1-core system, I think?
| TheRealPomax wrote:
| In the category "then update your FAQ, you've have many, many
| years to do so" =D
|
| (not to change the rules, but to explain why they rules
| _haven 't_ changed. Level playing fields are a worthwhile
| pursuit)
| caseyavila wrote:
| I do think it's interesting that recent submissions use nearly
| the entire 50 hours. I wonder how much better people could do
| if faster hardware was allowed.
| dang wrote:
| Related. Others?
|
| _Saurabh Kumar 's fast-cmix wins EUR5187 Hutter Prize Award_ -
| https://news.ycombinator.com/item?id=36839446 - July 2023 (1
| comment)
|
| _Hutter Prize Submission 2021a: STARLIT and cmix (2021)_ -
| https://news.ycombinator.com/item?id=36745104 - July 2023 (1
| comment)
|
| _Hutter Prize Entry: Saurabh Kumar 's "Fast Cmix" Starts 30 Day
| Comment Period_ - https://news.ycombinator.com/item?id=36154813 -
| June 2023 (5 comments)
|
| _Hutter Prize_ - https://news.ycombinator.com/item?id=33046194 -
| Oct 2022 (3 comments)
|
| _Hutter Prize_ - https://news.ycombinator.com/item?id=26562212 -
| March 2021 (48 comments)
|
| _500 '000EUR Prize for Compressing Human Knowledge_ -
| https://news.ycombinator.com/item?id=22431251 - Feb 2020 (1
| comment)
|
| _Hutter Prize expanded by a factor of 10_ -
| https://news.ycombinator.com/item?id=22388359 - Feb 2020 (2
| comments)
|
| _Hutter Prize: up to 50k EUR for the best compression algorithm_
| - https://news.ycombinator.com/item?id=21903594 - Dec 2019 (2
| comments)
|
| _Hutter Prize: Compress a 100MB file to less than the current
| record of 16 MB_ - https://news.ycombinator.com/item?id=20669827
| - Aug 2019 (101 comments)
|
| _New Hutter Prize submission - 8 years since previous winner_ -
| https://news.ycombinator.com/item?id=14478373 - June 2017 (1
| comment)
|
| _Hutter Prize for Compressing Human Knowledge_ -
| https://news.ycombinator.com/item?id=7405129 - March 2014 (24
| comments)
|
| _Build a human-level AI by compressing Wikipedia_ -
| https://news.ycombinator.com/item?id=143704 - March 2008 (4
| comments)
| slashdev wrote:
| I think the mistake here is to require lossless compression.
|
| Humans and LLMs only do lossy compression. I think lossy
| compression might be more critical to intelligence. The ability
| to forget, change your synapses or weights, is crucial to being
| able to adapt to change.
| version_five wrote:
| Yeah it makes no sense to say it's inspired by intelligence and
| then require lossless which is definitionally rote work and not
| intelligent.
| whimsicalism wrote:
| Not true, a smart model could be really good at lossy
| compression and then you only have to store a small delta to
| make it lossless.
| ClassyJacket wrote:
| I'm no mathematician but I don't believe this is true.
| Lossless information encoding requires _all_ the original
| information to be present.
| AnotherGoodName wrote:
| Arithmetic coding allows you to make a prediction and
| only provide bits for correction.
|
| Have the de-compressor predict the next data based on the
| outcome so far (a statistical prediction of next data
| will be lossy as it won't always be correct). If the
| prediction is correct you need to spend very little to
| confirm that. If it's incorrect you'll need to spend data
| to correct it. Arithmetic coding is the best way to make
| this work.
|
| It's also been used by all winning entries of the Hutter
| prize so far.
| glitchc wrote:
| Or at least reproducible. It could still be compressed.
| vladf wrote:
| What
| AnotherGoodName wrote:
| That's literally arithmetic coding which is used by all
| winning entries in the above so far.
| sytelus wrote:
| Humans can do lossy or lossless. There are plenty of people who
| can recite entire Bible or Koran flawlessly.
| kadoban wrote:
| That's true, but it seems unlikely that that's a particularly
| important part of intelligence. The vast majority of people
| do _not_ do that type of memorization, are they still
| intelligent?
| anonylizard wrote:
| Many can recite the Koran flawlessly, its short and heavily
| encouraged in education through rote repetition.
|
| Much, much fewer can recite the bible, its many times longer.
|
| LLMs can also recite the bible and Koran flawlessly, given
| how frequent the text appears in their training material.
| TheRealPomax wrote:
| This is more the equivalent of asking humans to create an
| exact copy of the text, typesetting and all, including the
| publishing information, page numbers, and exact linebreaks.
| Not just recite the text, which would be a lossy encoding of
| the original.
|
| Humans are _terrible_ at lossless encoding of information, it
| 's what we invented machines for =D
| Supply5411 wrote:
| And there are humans that can jump 8ft in the air. Doesn't
| mean it's correct to say that "humans can jump 8ft in the
| air." Very few people are regurgitating verbatim information.
| mik1998 wrote:
| Lossy text compression has little utility.
| JumpCrisscross wrote:
| > _Lossy text compression has little utility_
|
| You're describing every book you've ever read and learned
| from.
| TheAlchemist wrote:
| I mean, come on man. For some reason, the nerd in me sees this
| and immediately adds it on my 'I really need to do this' list.
|
| Just memories of old times doing some similar (albeit less
| challenging probably) competitions on TopCoder almost a decade
| ago, and also the curiosity to see how I would manage it know,
| with experience. Given that the current scores are also very far
| from what they estimate the lower bound to be, this is really
| interesting ! The prize is however very misleading - per their
| own FAQ - the total possible payout is ~223k euros.
|
| Definitely not thanking you for the hours I will put into this !
| omoikane wrote:
| 500000 EUR is the prize pool. Each winner has to gain at least 1%
| improvement over previous record to claim a prize that is
| proportional to the improvement. Getting the full 500000 EUR
| prize requires an 100% improvement (i.e. compressing 1GB to zero
| bytes).
| lainga wrote:
| Ah... I had professors who graded like that
| phobotics wrote:
| Does it or does it just require 1% improvement over the last
| winner? As opposed to a static additional 1% improvement vs the
| initial best "score".
| omoikane wrote:
| It's 1% over the last winner. The latest winner has a total
| size of 114156155, compared to previous winner of 115352938.
| The payout was 500000 * (1 - 114156155 /
| 115352938) = 5187
|
| (see table near "Baseline Enwik9 and Previous Records
| Enwik8")
| bigyikes wrote:
| Probably if you succeed at this, 500,000 will be worthless to
| you
| sytelus wrote:
| Why? How does this improvement translates to more financial
| gains?
| Eduard wrote:
| because with that knowledge, you will be able to decompress
| 0 dollar to infinite dollars which the storage mafia will
| pay you for not publishing your breakthrough in making them
| obsolete.
___________________________________________________________________
(page generated 2023-09-13 23:00 UTC) |