alt Hacker NewsFunnily enough, now that I've gone back and reread the LLM's explanations, I've decided that point 1, the one you were least critical of, is garbage whereas points 2 and 3 are fine.
1. Mice usually clear drugs faster, not slower, than humans, so either point 1 is wrong or I'm missing something, and either way it's a bad explanation.
2. This point is fine. The use of the phrase "tumor architecture" in this context is common, for example this random paper https://www.cell.com/cancer-cell/fulltext/S1535-6108(12)0008... and several papers cited by the Nature review. I don't get your problem with the phrase "highly pressurized cores" is, or what you're calling a non-sequitur.
Maybe you're arguing that it's an oversimplification to imply that xenografts are just not dense or pressurized enough, and it would be better to emphasize that tumor microenvironments affect drug delivery and aren't accurately modeled, which... sure, I suppose, though it seems like a nitpick.
3. Come on, you can't possibly think this is a valid criticism, and not just a thing you made up to have something to say.
> I continue to assert that the LLM's badly-plagiarising some papers, lecture notes and/or textbooks, blended with bad pop-sci analogies to the point of incoherence.
Then please, strive to do better!
I mean that in earnest, not just as an insult. You hate reading bullshit? Me too. If you're not familiar with the term "tumor architecture", it takes five seconds to put it into Google Scholar before you start insisting it's made up. Reducing the amount of bullshit on this site is everyone's duty.
> Mice usually clear drugs faster, not slower, than humans
… I knew that. That was one of the few things here I knew. But I read the LLM explanation and it looked right, so I started questioning other things instead, and got myself confused enough about basic anatomy that I didn't realise I was confused. (At one point, I decided that "blood goes through the liver and kidneys at the same rate as it goes through the heart" was a reasonable approximation, which is obviously false.)
And this despite that I was specifically watching out for LLM bullshit, and trying my hardest not to believe it. I guess this is evidence for my claim that LLMs are a terrible way for non-experts to learn about a topic, but wow, I was not playing the role in this argument that I thought I was.
> Maybe you're arguing that it's an oversimplification
Nope. That sounds like my genre of pedantry, but I didn't (and don't) understand the topic well enough to make that argument, so I wasn't. I was arguing that the description paints a picture of something unrealistic.
> Come on, you can't possibly think this is a valid criticism,
It's the criticism I fact-checked most thoroughly! So… yes, I just made it up to have something to say. (Honestly, the "just inject it near the tumour site" thing is extremely dubious, too: that would only work if you somehow eliminated blood flow through the tumour.)
> If you're not familiar with the term "tumor architecture", it takes five seconds to put it into Google Scholar
I put it into Internet Archive Scholar, forgot the quotes, and it just showed me machine-learning papers; so since I'd never heard of it before, I assumed the term was was made up. Lesson learned. (The term is much rarer than "tumour microenvironment" in the literature, but it does appear once on the Wikipedia page for tumor microenvironment, so I'm not sure how I missed it.) I can't figure out what the term actually means, but it does clearly mean something, and has done since at least 1971 (and probably earlier). doi:10.1136/bjo.78.11.871 sort of explains, but not really.
You mentioned textbooks in another comment. Do you have a textbook recommendation? I think I clearly need remedial study.