The AI Thread

[Samson]

I have found what is to me the only useful example of AI copyright theft. What does CFC think about the morality?

Google have thrown a LLM at their google scholar tool. From a quick test it basically gives you the same list of papers the non-AI version but it has access to the whole paper for paywalled ones, and gives much more information that the abstract for many of them. Sure you need to get the real paper before you actually take any of it as fact, but it really reduces the number of papers you need to do that for. For paywalled papers not on sci hub that makes a massive difference to the accessibility of the information.

It is possibly hurting the publishers, but screw elisvier and their 40% profit margin on other peoples work. For the actual researchers it should help the metric they care about, references, as it makes their work more accessible.

 

[Gori the Grey]

It doesn't seem to me that it does make the papers more accessible.

This little write-up itself says that, to make any use of the essays generated, a researcher would have to go track down the real paper (by the old bibliographic methods).

All this seems to give you is whatever bibliography is produced by google scholar, the abstract of the paper, and then a summary composed by AI.

I doubt any researcher worth his or her salt would trust the summary composed by AI, so it's back to tracking down the original article, if it sounds promising by title or abstract.

I don't know how elisvier works and how they make a profit on scholarly articles. But whatever the problem is there, it doesn't seem to me that this fixes that. It might need to be addressed separately.

 

[Samson]

It doesn't seem to me that it does make the papers more accessible.

This little write-up itself says that, to make any use of the essays generated, a researcher would have to go track down the real paper (by the old bibliographic methods).

All this seems to give you is whatever bibliography is produced by google scholar, the abstract of the paper, and then a summary composed by AI.

I doubt any researcher worth his or her salt would trust the summary composed by AI, so it's back to tracking down the original article.

I don't know how elisvier works and how they make a profit on scholarly articles. But whatever the problem is there, it doesn't seem to me that this fixes that. It might need to be addressed separately.

It allows you to triage the papers more effectively. You are not going to trust the summary to use it for a real answer, but it could easily be enough to either know you do not need it, and therefore not go to the effort of ordering it unnecessarily, or tell you that you really do want it, so you might go to the effort when you otherwise would not have bothered.

 

[Gori the Grey]

Not more than the author's abstract. I, at least, don't trust AI summaries to really get at what would be potentially important about an article.

 

[Samson]

Not more than the author's abstract. I, at least, don't trust AI summaries to really get at what would be potentially important about an article.

If the abstract is good you may well be right. There are too many abstracts that do not answer the important questions one at least sometimes need to know/