The "here is a 200 word summary; would you like me to make any changes?" line doesn't necessarily imply that one fabricated data, but it certainly implies that you didn't actually read what it spit out, which means the AI could easily have fabricated information and no one has checked it. So the whole thing becomes suspect. And if you did it once, should I trust your next paper where you remembered to delete the proof of your sloppiness any more?

As a reviewer, pre-AI, I have recommended rejection of a paper with multiple references that did not support the facts they were citing from them, on the grounds that the paper demonstrated a slipshod approach to research that made me uncomfortable trusting them on the core claims. It wasn't published, so it seems the editor likely agreed with me. Was anyone banned for it? Probably not, but it makes me think twice before I trust anything those authors write now.

They're not just competing for submissions, their entire value depends on the material on ArXiv having value as scientific literature. They need to protect that above all else, especially since ArXiv has fewer guardrails than a typical scientific journal.

I know the ship has sailed, but I don't think people should be using AI at all to write scientific journal articles. To do research, sure, to process data, absolutely, to find references, of course, but every word in the article should be written by a human who knows what they are trying to convey and who is responsible if it is wrong. And you shouldn't cite references you haven't read (at least the part your referencing). At very least, if you want to generate AI work, it's not too much to ask that you meticulously curate the results.

AI can only generate somewhat useful scientific work because it has a mostly trustworthy body of literature to pull from. If we let AI poison that body of literature, pretty soon both the literature and the AI will be worthless.

The study basically says they were unable to account for differences in cultivar for most crops, and in the supplemental they show some of the data for those they were able to look at by cultivar—and they don't look at the effect of changes in cultivar choice, they just look at whether nutritional values change within cultivar. They only present significant changes, which skews perceptions, and the direction of change is far from consistent.

The other factor is that while nutrient content may have gone down ~3% in the past four decades, food consumption has increased far more, especially in some parts of the developing world. So most people are getting more nutrients, not less, even if the nutrient density has fallen. There are certainly going to be people on the edges of the distributions that are going to be deficient, and we shouldn't dismiss that, but I still think that by and large that number is going to be smaller now and in the future than it was in the 80s.

These are a loss-of-function gene edit, not transgenic. They aren't expressing any foreign genes or producing any foreign proteins. They are just not making a protein they used to make—polyphenol oxidase. It may be making a truncated version of polyphenol oxidase (they don't seem to have made public yet where they edited the gene) but the odds are that nonsense mediated decay is breaking down the RNA transcript before much of the truncated protein is made.

1) This really is a non-issue for banana, which commercially is essentially all one cultivar, Dwarf Cavendish (there are other bananas at small scale in tropical countries, but a non-browning Dwarf Cavendish wouldn't fill the niches they fill now any better than browning Dwarf Cavendish does).

2) You can still plant the original Dwarf Cavendish if you don't want to pay royalties. And the protection doesn't last forever. Plus, honestly, one of the reasons more of this work isn't done in tropical crops is that it's going to be impossible to enforce these IP rights on small holder farms in developing countries. I think the odds of anyone coming after the guy with a few plants in his yard are vanishingly small.

My feeling on these is: Did we really need this? I might throw away a couple of bananas a year. And bananas are super cheap. The savings to me is a maximum of a few dollars, and maybe a little less banana bread. I mean, it's an upgrade, but would I pay significantly more for it?

Since we're doing anecdotes, I have a counter-anecdote.

I live in a rural residential kind of area—2-10 acre lots, mix of light woods, yards, barns and paddocks for horses. A few lots away, there was an old woman who would accept feral cats from all around the county, have them sterilized, and then release them from her house. I put up a game camera in the backyard and I eventually stopped looking at it regularly because it was just picture after picture of cats. Another neighbor who has lived there longer than I have mentioned that before she started doing that, the birds in the area were much more diverse. Five or six years ago the woman died, the steady stream of feral cats ended, and the return of diversity has been VERY noticeable. I've seen maybe a dozen bird species I never saw in the preceding seven years, and species which were once rare have become much more common. I've seen far more squirrels and lizards, maybe even more coyotes (presumably because of more prey...I think I would have noticed feral cats capable of eating coyotes). The quail, which were easy prey on the ground, have returned in force—I see them literally every day now, after years without seeing them at all.

Feral cat diet composition isn't unknown—there are multiple studies in multiple locations looking at species composition in stomach contents and feces, which are cited in the Nature article.

I totally agree. In fact, there's a very large problem in the other direction--an awful lot of people don't see how software/music/video, which is readily duplicated and lacking in physical substance, has a significant monetary value at all.