A great deal of effort and money is spent running studies. I'm inclined to assume the experts in the field are more aware of the tradeoffs of that decision and how to mitigate the downsides than probably all, but certainly the overwhelming majority, of people commenting on this thread.
Someone who needs to ask an LLM will not be helpful in trying to point out something they missed.
They're not pointing out something the researchers missed, they're pointing out something the people in this thread confidently hyping the results are missing. I'm certain the researchers are familiar with the limitations of the models they used (is it bad that the incentives of science and science journalism leads to overoptimistic coverage that hint at groundbreaking implications without explaining to lay readers what the unknowns are? Probably, but that's not these researchers' faults).
The average person in this thread, however, would probably be better informed by asking an LLM for context. They'd be even better informed by taking a few weeks to work through a textbook on cancer biology, but realistically they won't.
My horse in the race is that I'm annoyed by overenthusiastic comments that display a lack of understanding of the history of cancer treatment, and I'm going to be even more annoyed in a few months when the rounds of "haven't we had 1000 cures to cancer posted to HN??? why aren't we using any of them???" start showing up again. I'd rather encourage informed, skeptical optimism.
No, they aren't: the second is irrelevant and unphysical. Highly-pressurised cores? Really? "Dense", I could buy, but:
• If there's blood supply, then (A) it can't be a much higher pressure than the blood pressure (unless there's some Rube Goldberg machine involving active transport), and (B) the tumour is reachable by treatments like this;
• And if there isn't blood supply, then the tumour's core is necrotic, and a treatment to kill the dead cells wouldn't do anything anyway. (Sure, killing the tissue that isolates a lump of necrotic flesh from the rest of the body might cause new and exciting problems, but somehow I think those might be preferable to incurable breast cancer.)
The second is just not a relevant criticism. The third, if it's an actual issue, can probably be worked around by tweaking the molecule slightly – and if not, suppressing the immune system isn't that difficult (it's a known side-effect of many chemotherapies). The first, if it's an issue, can be avoided by injecting the medicine near the target site.
I agree that this treatment might not work in humans, but all the AI's done is taken a generic list of potential concerns, and inserted technobabble to try to make it match the scenario. If you want generic criticism, see https://news.ycombinator.com/item?id=47209076: at least that's true.
You're incredibly wrong. You also cited my own comment at me.
The problem of high interstitial pressure (not blood pressure) interfering with drug delivery in tumors is basic cancer biology. If you don't believe me, here's:
A review published in a reputable oncology journal, with over 100 citations, entirely about targeting interstitial pressure, with an abstract leading with "Tumor interstitial pressure is a fundamental feature of cancer biology. Elevation in tumor pressure affects the efficacy of cancer treatment."
https://aacrjournals.org/cancerres/article/74/10/2655/592612...
Another review, also a reputable oncology journal, 1000 citations, about tumor stroma more generally, which lists high interstitial pressure as a mechanism by which tumors limit drug access and includes a nice diagram (Figure 2a).
https://www.nature.com/articles/s41571-018-0007-1
That's how basic this fact is. 1000 citation reviews in Nature have beautiful fucking diagrams of it. I'm pretty sure it was in the textbook of my undergraduate biology class.
If you don't know shit, don't talk shit. People will criticize LLMs for being overconfident while writing essays from their ass.
I did briefly consider that this was referring to intercellular fluid, but "highly pressurized cores" is a terrible way to describe high IFP, so I rejected that interpretation. I thought the LLM was "trying to" refer to some kind of dense-walled cyst. (Of course, the LLM wasn't actually trying to say anything at all.) (And I think my argument there about osmotic pressure, oxygen diffusion and tissue necrosis is correct: hypoxia's already an issue for tumours, and there are only so many heroic workarounds available before a cell's only option is to die; and since blood pressure is higher than even the high IFP found in tumours, that's the appropriate bound for the argument I made.)
Your interpretation leaves the LLM's discussion of stroma as a non-sequitur, since that is not why high IFP causes problems for drug uptake; and at that point, I think you're just substituting a correct statement in place of the LLM's superficially-meaningful nonsense. I'll go through it again, this time focusing on the names assigned to each point:
"1. The Scale of the Human Body" talks about the excretory system. The part of the explanation comparing "tiny" to "vast" is at the very least misleading, but I would call it outright wrong. And yes: I am also thinking of all those "well actually, the geometry of the circulatory system" interpretations that make it technically correct, but… if a biology teacher explained it like this, would you really think they were teaching it properly? (I mean, seriously, calling "The human liver, spleen, and kidneys" "the reticuloendothelial system"‽)
"2. Tumor Architecture", under your charitable interpretation, isn't talking about "architecture" at all.
"3. Immune System Differences" is at least named right; but a treatment that only works in immunosuppressed patients is still a treatment. You could imagine a cancer drug sufficiently-effective that it is worth suppressing the immune system just so you can administer it. (I don't think it's likely that this is one, but that's for experiment to decide. And if the patient's immunocompromised anyway…)
> You also cited my own comment at me.
Oops. That does make me feel foolish. In my defence: it didn't occur to me that anyone could think you were saying the same things as the LLM, because what you were saying was correct, and what the LLM was saying was nonsense.
Although, if you thought you were saying the same thing… is the LLM's "tumor architecture" supposed to refer to the tumour microenvironment‽ That would explain the stroma mention, but… wow that is not a sensible way to say that. 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.
A great deal of effort and money is spent running studies. I'm inclined to assume the experts in the field are more aware of the tradeoffs of that decision and how to mitigate the downsides than probably all, but certainly the overwhelming majority, of people commenting on this thread.
Someone who needs to ask an LLM will not be helpful in trying to point out something they missed.
They're not pointing out something the researchers missed, they're pointing out something the people in this thread confidently hyping the results are missing. I'm certain the researchers are familiar with the limitations of the models they used (is it bad that the incentives of science and science journalism leads to overoptimistic coverage that hint at groundbreaking implications without explaining to lay readers what the unknowns are? Probably, but that's not these researchers' faults).
The average person in this thread, however, would probably be better informed by asking an LLM for context. They'd be even better informed by taking a few weeks to work through a textbook on cancer biology, but realistically they won't.
My horse in the race is that I'm annoyed by overenthusiastic comments that display a lack of understanding of the history of cancer treatment, and I'm going to be even more annoyed in a few months when the rounds of "haven't we had 1000 cures to cancer posted to HN??? why aren't we using any of them???" start showing up again. I'd rather encourage informed, skeptical optimism.
No, they aren't: the second is irrelevant and unphysical. Highly-pressurised cores? Really? "Dense", I could buy, but:
• If there's blood supply, then (A) it can't be a much higher pressure than the blood pressure (unless there's some Rube Goldberg machine involving active transport), and (B) the tumour is reachable by treatments like this;
• And if there isn't blood supply, then the tumour's core is necrotic, and a treatment to kill the dead cells wouldn't do anything anyway. (Sure, killing the tissue that isolates a lump of necrotic flesh from the rest of the body might cause new and exciting problems, but somehow I think those might be preferable to incurable breast cancer.)
The second is just not a relevant criticism. The third, if it's an actual issue, can probably be worked around by tweaking the molecule slightly – and if not, suppressing the immune system isn't that difficult (it's a known side-effect of many chemotherapies). The first, if it's an issue, can be avoided by injecting the medicine near the target site.
I agree that this treatment might not work in humans, but all the AI's done is taken a generic list of potential concerns, and inserted technobabble to try to make it match the scenario. If you want generic criticism, see https://news.ycombinator.com/item?id=47209076: at least that's true.
You're incredibly wrong. You also cited my own comment at me.
The problem of high interstitial pressure (not blood pressure) interfering with drug delivery in tumors is basic cancer biology. If you don't believe me, here's:
A review published in a reputable oncology journal, with over 100 citations, entirely about targeting interstitial pressure, with an abstract leading with "Tumor interstitial pressure is a fundamental feature of cancer biology. Elevation in tumor pressure affects the efficacy of cancer treatment." https://aacrjournals.org/cancerres/article/74/10/2655/592612...
Another review, also a reputable oncology journal, 1000 citations, about tumor stroma more generally, which lists high interstitial pressure as a mechanism by which tumors limit drug access and includes a nice diagram (Figure 2a). https://www.nature.com/articles/s41571-018-0007-1
That's how basic this fact is. 1000 citation reviews in Nature have beautiful fucking diagrams of it. I'm pretty sure it was in the textbook of my undergraduate biology class.
If you don't know shit, don't talk shit. People will criticize LLMs for being overconfident while writing essays from their ass.
I did briefly consider that this was referring to intercellular fluid, but "highly pressurized cores" is a terrible way to describe high IFP, so I rejected that interpretation. I thought the LLM was "trying to" refer to some kind of dense-walled cyst. (Of course, the LLM wasn't actually trying to say anything at all.) (And I think my argument there about osmotic pressure, oxygen diffusion and tissue necrosis is correct: hypoxia's already an issue for tumours, and there are only so many heroic workarounds available before a cell's only option is to die; and since blood pressure is higher than even the high IFP found in tumours, that's the appropriate bound for the argument I made.)
Your interpretation leaves the LLM's discussion of stroma as a non-sequitur, since that is not why high IFP causes problems for drug uptake; and at that point, I think you're just substituting a correct statement in place of the LLM's superficially-meaningful nonsense. I'll go through it again, this time focusing on the names assigned to each point:
"1. The Scale of the Human Body" talks about the excretory system. The part of the explanation comparing "tiny" to "vast" is at the very least misleading, but I would call it outright wrong. And yes: I am also thinking of all those "well actually, the geometry of the circulatory system" interpretations that make it technically correct, but… if a biology teacher explained it like this, would you really think they were teaching it properly? (I mean, seriously, calling "The human liver, spleen, and kidneys" "the reticuloendothelial system"‽)
"2. Tumor Architecture", under your charitable interpretation, isn't talking about "architecture" at all.
"3. Immune System Differences" is at least named right; but a treatment that only works in immunosuppressed patients is still a treatment. You could imagine a cancer drug sufficiently-effective that it is worth suppressing the immune system just so you can administer it. (I don't think it's likely that this is one, but that's for experiment to decide. And if the patient's immunocompromised anyway…)
> You also cited my own comment at me.
Oops. That does make me feel foolish. In my defence: it didn't occur to me that anyone could think you were saying the same things as the LLM, because what you were saying was correct, and what the LLM was saying was nonsense.
Although, if you thought you were saying the same thing… is the LLM's "tumor architecture" supposed to refer to the tumour microenvironment‽ That would explain the stroma mention, but… wow that is not a sensible way to say that. 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.
1 reply →