The official blog of arXiv, the preprint repository, announcing site features, infrastructure changes, and submission policies.
HN links it for arXiv's HTML paper format, its Fastly CDN move, and its policy on ChatGPT-style tools, which commenters debate in terms of cost and format.
I‘m not convinced this exercise in what to and what not to include in this cost-benefit-analysis will lead to anything. We can always arbitrarily include an extra item to include to shift the calculations in our favor. For example I could simply add the cost of creating the data which is fed into the training set of an LLM, that creation is done by our human biological machinery and hence has the cost of the frozen blueberries, the rigid fiber insulations, the machinery that dug the waterpipe for their shower, etc.
Instead I would like to shift the focus on the benefits of LLM. I know the costs are high, very very very high, but you seem to think that the benefits are also so high measured in time saved. That is the amount of tasks automated are enough to save humans doing similar tasks by miles. If that is what you think I disagree. LLMs have yet to prove them selves with real world application. We are seeing when we actually do measure how much LLMs save work-hours, that it the effects are at best negligible (see e.g. https://news.ycombinator.com/item?id=44522772). Worse, generative AI is disrupting our systems in worse way, where e.g. teachers, peer-reviewers, etc. have to put in a bunch of extra work to verify that the submitted work was actually written by that person, and not simply generated by AI. Just last Friday I read that arXiv will no longer accept submissions unless they have been previously peer-reviewed because they are overwhelmed by AI generated…
arXiv doesn't need much. All they do is host static pdfs uploaded by someone else with free CDN services from Fastly [0]. I'm sure they could get academics to volunteer moderation services as well.
In reality you could host the entire thing for well under $50k/year in hardware and storage if someone else is providing a free CDN. Their costs could be incredibly low.
But just like Wikipedia I see them very likely very quickly becoming a money hole that pretends to barely be kept afloat from donations. All when in reality whats actually happening is that its a ridiculous number of rent seekers managed to ride the coattails of being the defacto preprint server for AI papers to land themselves cushy Jobs at a place that spends 90+% of their money on flights and hotels and wages for their staff.
I'm already expecting their financial reports to look ridiculously headcount heavy with Personnel Expenses, Meetings and Travel blowing up. As well as the classic Wikipedia style we spend a ton of money in unclear costs [1].
Whats already sad is they stopped having a real broken down report that used to actually showed things. Like look at this beautiful screenshot of a excel sheet. Imagine if Wikipedia produced anything this clear. [2]
[0] https://blog.arxiv.org/2023/12/18/faster-arxiv-with-fastly/
[1] https://info.arxiv.org/about/reports/FY26_Budget_Public.pdf
[2] https://info.arxiv.org/about/reports/2020_arXiv_Budget.pdf
It would already be a step in the right direction, if papers would also publish a VM with all their code, data and dependencies.
It is nice to have the code (https://blog.arxiv.org/2020/10/08/new-arxivlabs-feature-provides-instant-access-to-code/), but without necessary dependencies, the correct OS, compiler Version, etc. replication is even with code often impossible.
Having running demos is another step in the right direction (see https://blog.arxiv.org/2022/11/17/discover-state-of-the-art-machine-learning-demos-on-arxiv/).
But outside of computer science replication is even more difficult. Maybe if people would use standardized laboratories and robots, one could replicate findings by rerunning the robots code on another standard robot lab ( Basically the idea here is to virtualize laboratory work).
But even then for the biggest most complex experiments this will not work: Replicate CERN anyone?
You may reasonably disagree with my comment, but using your imagination you'll find that there are lots of ways by which arXiv (the organization) and its staff could benefit from PR.
Lots of academics in distant fields are unaware of arXiv, and even academics (like me) who use arXiv daily and host their preprints there don't think of it as any more than a place to store, catalogue, and retrieve papers.
Look at all the ways in which arXiv is (like any institution) perpetuating and expanding itself: https://blog.arxiv.org/
Look at these extra things arXiv is doing (including commercial integrations): https://info.arxiv.org/labs/showcase.html
I've been gratefully using and contributing to arXiv since 2008, and I hope it continues to be the incredible resource that it is. I think your take is naive and that even great institutions can end up like Mozilla.
Hi, ar5iv creator here. Just to jot down that:
1. This was first reported to HN 55 days ago, discussion is here:
https://news.ycombinator.com/item?id=30163308
2. There is an official arXiv blog announcement with a little bit of extra details here:
https://blog.arxiv.org/2022/02/21/arxiv-articles-as-responsive-web-pages/
---
When I first commented on HN, I mentioned that I really hoped we can find a way for arXiv to officially endorse the project and adopt it, so that we solve the copyright issues and get on a healthy path to full integration with the service. That happened a month ago! More healthy work ahead.
Thanks to everyone on HN for the feedback so far. If/when you see broken renderings (still plenty of examples) do comment with the article link, or (even better) use the "report an issue" button at the bottom of each document page. Cheers!
These features all sound great, but it's still extending PDF "upwards" from a quite low-level, representation-focused format.
For most content PDFs are used today, I'd much prefer extending something like ePub "downwards" with stylesheets, rendering hints etc. – that can all just be thrown out by an accessibility- or readability-focused viewer.
Yet, the PDF ecosystem is so large at this point, and so many organizations are still tied to skeuomorphic ideas focused around printed pages of text, that I can't see it happen anytime soon. I just hope that at least academia figures something out.
ArXiv's recent HTML efforts [1] are a great step in that direction, but as I understand it, they still use PDF as their main ingestion format.
[1] https://blog.arxiv.org/2023/12/21/accessibility-update-arxiv-now-offers-papers-in-html-format/