I (economist) use Google Scholar less and less. That's partly from aversion to Google in general as well as the specific experience of being burned by abandoned Google products, as you mention. But more because the alternatives, including AI literature summary tools, are getting better and better.
For arxiv, we have RePeC, though this is pretty much a one man show.
I concur about Google abandoning products. Their cloud database was an example. It is a reminder never to be reliant on a single vendor/platform that can pull the rug out from under a product built on top. Bricking, or crippling products, depending on a vendor's service, is now a thing. The latest is Google's removal of connectivity to the Nest thermostat. The same goes for Amazon, and selling products in its marketplace.
The "universal archive" role of arxiv and its (ideally) permanent links are critical for many open-access journals that act as overlays on arxiv. Maintaining this focus seems critically useful for the field.
ArXiv was set up with the main filter being the technical one of being able to produce latex source files. It was also set up without a linking or commenting option, which seems to be a larger drawback. As a result of the first, there are a significant numbers of papers that have been posted there that are completely worthless and wrong: For example, there continues to be a steady rate of complete garbage P=NP and P!=NP papers posted on arxiv - 20 years ago, these papers actually formed a large proportion of all submissions under certain subject identifiers. Without a linking or commenting feature how can the community let others know what the problems are? The partial solution that some groups have found is to produce other arXiv papers with refutations of these bogus claims, but there is no way of adding "cited by" links through the system so someone reading the original garbage will see them.
The issue that arxiv has with review articles is clearly one of the rate of garbage production and the costs of serving that garbage rather than the fact that there is garbage. The "review article" limitation might be the current expedient, but it doesn't seem to work longer term as people use agents to swamp the system with other kinds of useless papers. Absent linking/commenting, maybe arxiv could avoid a ban and use levels of serving - the analog of an ordinary library putting some books that nobody seems to have a use for in the basement stacks - to reduce the costs.
Just remember the mantra: "Moderation cannot be done well at scale". One can moderate a website or blog, but not well on platforms, e.g., Facebook and X/Twitter. Trying leads to shortcuts, automation of moderation, countered by gaming with AI slop, bots, and troll farms. It is like malware attacks. Large user base OSs and code libraries attract malware as the potential spread of teh vector is large. An OS or niche code base that has a small user base will be ignored.
Perhaps the better bet is to set up a separate, but linked site that allows reviews and similar, with the caveat that the content will not be well-moderated, if at all, but keeps the main journal site free of garbage. IDK whether AI-generated papers are slipping through on arXiv, but Sabine Hoffstader may have reported on at least one, IIRC (or was it just plain human-generated garbage?).
As someone who doesn't use arxiv, but does use other preprint servers that have already banned/disallowed reviews (eg biorxiv, medrxiv), I guess I don't see what all the fuss is about. If folks want to publish a review article, why not send it to a peer reviewed journal? It's not like folks are locked out of sharing their opinions.
I think I was pretty clear on this point. Why should "research" papers be prioritized by preprint servers over "survey" papers? This is a community value judgment.
I (economist) use Google Scholar less and less. That's partly from aversion to Google in general as well as the specific experience of being burned by abandoned Google products, as you mention. But more because the alternatives, including AI literature summary tools, are getting better and better.
For arxiv, we have RePeC, though this is pretty much a one man show.
I concur about Google abandoning products. Their cloud database was an example. It is a reminder never to be reliant on a single vendor/platform that can pull the rug out from under a product built on top. Bricking, or crippling products, depending on a vendor's service, is now a thing. The latest is Google's removal of connectivity to the Nest thermostat. The same goes for Amazon, and selling products in its marketplace.
The "universal archive" role of arxiv and its (ideally) permanent links are critical for many open-access journals that act as overlays on arxiv. Maintaining this focus seems critically useful for the field.
ArXiv was set up with the main filter being the technical one of being able to produce latex source files. It was also set up without a linking or commenting option, which seems to be a larger drawback. As a result of the first, there are a significant numbers of papers that have been posted there that are completely worthless and wrong: For example, there continues to be a steady rate of complete garbage P=NP and P!=NP papers posted on arxiv - 20 years ago, these papers actually formed a large proportion of all submissions under certain subject identifiers. Without a linking or commenting feature how can the community let others know what the problems are? The partial solution that some groups have found is to produce other arXiv papers with refutations of these bogus claims, but there is no way of adding "cited by" links through the system so someone reading the original garbage will see them.
The issue that arxiv has with review articles is clearly one of the rate of garbage production and the costs of serving that garbage rather than the fact that there is garbage. The "review article" limitation might be the current expedient, but it doesn't seem to work longer term as people use agents to swamp the system with other kinds of useless papers. Absent linking/commenting, maybe arxiv could avoid a ban and use levels of serving - the analog of an ordinary library putting some books that nobody seems to have a use for in the basement stacks - to reduce the costs.
Just remember the mantra: "Moderation cannot be done well at scale". One can moderate a website or blog, but not well on platforms, e.g., Facebook and X/Twitter. Trying leads to shortcuts, automation of moderation, countered by gaming with AI slop, bots, and troll farms. It is like malware attacks. Large user base OSs and code libraries attract malware as the potential spread of teh vector is large. An OS or niche code base that has a small user base will be ignored.
Perhaps the better bet is to set up a separate, but linked site that allows reviews and similar, with the caveat that the content will not be well-moderated, if at all, but keeps the main journal site free of garbage. IDK whether AI-generated papers are slipping through on arXiv, but Sabine Hoffstader may have reported on at least one, IIRC (or was it just plain human-generated garbage?).
As someone who doesn't use arxiv, but does use other preprint servers that have already banned/disallowed reviews (eg biorxiv, medrxiv), I guess I don't see what all the fuss is about. If folks want to publish a review article, why not send it to a peer reviewed journal? It's not like folks are locked out of sharing their opinions.
I think I was pretty clear on this point. Why should "research" papers be prioritized by preprint servers over "survey" papers? This is a community value judgment.