Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Arxiv.org on IPFS

xirva.org

Post reply on HN
Print view View original

61–70 of 77 posts

Re: Show HN: Arxiv.org on IPFS

#61

Post by amelius »

Taek wrote: post #43
imback wrote: post #38

Slightly OT, but how is IPFS these days? Every time I've tried using it the past few years it seemed very WIP. Cool sounding protocol though.

It's still very slow - retrieval times >30s for files that aren't cached on cloudflare or ipfs.io. Also, those two providers each have multiple periods of downtime every year. If the files are cached and the services are up, it's plenty fast for static data but dynamic data (IPNS) is still very slow. We've built a competitor called Skynet that is much faster (less than 200ms for files that aren't in the cache) and sc…

Why do you need a blockchain to implement Skynet?

Re: Show HN: Arxiv.org on IPFS

#62

Post by elcritch »

hugoroussel wrote: post #31
lmohseni wrote: post #29

Earlier quoted context omitted.

I believe science is a democratic process. If someone has the correct idea but communicates it poorly, so poorly that others in the field disagree, then this person is doing it wrong. (Thinking specifically of https://en.m.wikipedia.org/wiki/Shinichi_Mochizuki )

All and all it is a non trivial problem. You have at the very least have to attach some kind of form of reputation system into the verification process. Even with that you will still have the "misunderstood genius" issue, or the "excellent reputation professor" that everyone trust without (enough) verification.

But at least there’s be a system for other researchers to record “failed to replicate” that could give a channel to critique reputable professors that’s not controlled by the same professors (as they often can in journals).

Re: Show HN: Arxiv.org on IPFS

#63

Post by SkyMarshal »

Taek wrote: post #50
onhn wrote: post #18

"You can upload your research and publish it on the open web. Members of the community will be able to vote on your research to raise its visibility." Oh dear.

How would you set it up? The decentralized world doesn't really have a great system for curation at this point (unless you can point to a counterexample!), and so I'm in favor of any sort of playing around with decentralized voting/curation until we find something that seems to be working well.

Start from the objective of first do no harm. Voting systems may eventually be gamed to distort results, so eliminate the voting system. Instead rely on ad-hoc personal networks to disseminate signal about quality papers out-of-band. Don’t assume you have to systematize everything.

Re: Show HN: Arxiv.org on IPFS

#64

Post by Taek »

pimterry wrote: post #54
Taek wrote: post #43

Earlier quoted context omitted.

It's still very slow - retrieval times >30s for files that aren't cached on cloudflare or ipfs.io. Also, those two providers each have multiple periods of downtime every year. If the files are cached and the services are up, it's plenty fast for static data but dynamic data (IPNS) is still very slow. We've built a competitor called Skynet that is much faster (less than 200ms for files that aren't in the cache) and sc…

Do you have any documentation that compares Skynet to IPFS anywhere? From an architectural standpoint and/or features.

Closest we got at the moment is this: https://skynet.guide/discover/storage-chains-compared.html#f...

I'm writing a more direct comparison this week, we just recently (less than 30 days ago) hit full feature parity with ipfs, any webapp or file deployed on IPFS should work natively on Skynet now as well. The link/identifier will be different but you shouldn't need to change any code

Re: Show HN: Arxiv.org on IPFS

#65

Post by Taek »

amelius wrote: post #61
Taek wrote: post #43

Earlier quoted context omitted.

It's still very slow - retrieval times >30s for files that aren't cached on cloudflare or ipfs.io. Also, those two providers each have multiple periods of downtime every year. If the files are cached and the services are up, it's plenty fast for static data but dynamic data (IPNS) is still very slow. We've built a competitor called Skynet that is much faster (less than 200ms for files that aren't in the cache) and sc…

Why do you need a blockchain to implement Skynet?

The blockchain gets us a decentralized marketplace for decentralized storage providers. Anyone can join as a provider and get paid, and the blockchain can act as a decentralized escrow that holds the payment until proof is provided that the storage contract was properly fulfilled.

98% of our technology is off-chain. Only a little tiny sliver (the file contract open and close) is actually posted to the blockchain.

Re: Show HN: Arxiv.org on IPFS

#66

Post by nextaccountic »

lmohseni wrote: post #29
NmAmDa wrote: post #28

Earlier quoted context omitted.

Actually I don't think science has democratic nature. Yes we do somehow do that as a theory would still need to be accepted widely. But in reality one person can have the correct idea while all others disagree. Still this person is doing it right.

I believe science is a democratic process. If someone has the correct idea but communicates it poorly, so poorly that others in the field disagree, then this person is doing it wrong. (Thinking specifically of https://en.m.wikipedia.org/wiki/Shinichi_Mochizuki )

The participation ought to be democratic in the sense of being open to everyone to participate. But, you can't do a vote and use it to decide who is right. Deep down we know that being right or wrong is independent from the scientific consensus. Mochizuki may be interacting with the scientific community in the wrong way, but it has no bearing on whether his theory is correct.

The consensus itself has some democratic features, but it's weighed by prestige and adherence to the current paradigm. I think Kuhn described its mechanism pretty well. It's far easier to convince people of a wrong result if you follow the established paradigm, than convince people of something right if you go against it. What really saves science from being pure dogma is that there are paradigm shifts, revolutions in which the scientific consensus change.

Re: Show HN: Arxiv.org on IPFS

#67

Post by meroje »

hevalon wrote: post #7

Just a heads up comment, it seems some of the IPFS links are broken. For example I just visited https://www.xirva.org/categories/cs.AI and all links point to an "undefined" path example https://ipfs.io/ipfs/undefined/2107.00082.pdf But overall, great idea!

links ? everything is a fake js button

Re: Show HN: Arxiv.org on IPFS

#68

Post by a1369209993 »

Taek wrote: post #51
menmob wrote: post #48

Earlier quoted context omitted.

Are you able to “pin” files yourself like IPFS? Or are you required to pay hosts to store it?

You can always run a host yourself and pin it to your own host. The main reason we chose a host-based architecture instead of a pin based architecture is that we saw on IPFS that having people pin their own data resulted in really poor uptimes, a lot of file rot, and it also substantially reduced scalability and increased fetch times. And after all of those tradeoffs, the vast majority of accessible content on IPFS i…

> You can always run a host yourself and pin it to your own host.

(It sounds like it, but) to clarify, can you do this completely for free, with no cooperation from a third party (eg, you don't need to pay a existing host to vouch for you)?

Re: Show HN: Arxiv.org on IPFS

#70

Post by Taek »

a1369209993 wrote: post #68
Taek wrote: post #51

Earlier quoted context omitted.

You can always run a host yourself and pin it to your own host. The main reason we chose a host-based architecture instead of a pin based architecture is that we saw on IPFS that having people pin their own data resulted in really poor uptimes, a lot of file rot, and it also substantially reduced scalability and increased fetch times. And after all of those tradeoffs, the vast majority of accessible content on IPFS i…

> You can always run a host yourself and pin it to your own host. (It sounds like it, but) to clarify, can you do this completely for free, with no cooperation from a third party (eg, you don't need to pay a existing host to vouch for you)?

No need to pay an existing host or get any sort of external party to enable you, it's a permissionless network
Post reply on HN