Haotian's Shortform
post by Haotian (haotian-huang) · 2024-12-12T22:16:57.820Z · LW · GW · 9 commentsContents
9 comments
9 comments
Comments sorted by top scores.
comment by Haotian (haotian-huang) · 2024-12-12T22:16:58.016Z · LW(p) · GW(p)
LessWrong, is this rational? I wrote a reply to Elizabeth's open bid for answers to her research questions in good faith. She replies not with anything substantive, but to claim it is written with AI. I'm happy for her to disagree with my answer, but flagging a difference in style to suggest low quality is not what I thought this community was supposed to be about. Cynically, one could suggest she doesn't want to make good on her offer ... For the record, I wrote it late at night and ran the response through an LLM to improve readability for her benefit.
If you don't know what I'm talking about, see the most disliked comment on her post :)
https://www.lesswrong.com/posts/bFvEpE4atK4pPEirx/correct-my-h5n1-research-usdreward [LW · GW]
Replies from: cata, habryka4, gwern, Larks↑ comment by cata · 2024-12-12T22:59:54.727Z · LW(p) · GW(p)
To me, since LessWrong has a smart community that attracts people with high standards and integrity, by default if you (a median LW commenter) write your considered opinion about something, I take that very seriously and assume that it's much, much more likely to be useful than an LLM's opinion.
So if you post a comment that looks like an LLM wrote it, and you don't explain which parts were the LLM's opinion and which parts were your opinion, then that makes it difficult to use it. And if there's a norm of posting comments that are partly unmarked LLM opinions, that means that I have to adopt the very large burden of evaluating every comment to try to figure out whether it's an LLM, in order to figure out if I should take it seriously.
Replies from: haotian-huang↑ comment by Haotian (haotian-huang) · 2024-12-13T05:18:15.718Z · LW(p) · GW(p)
Thank you for your comment. I will highlight specifically which parts are my opinion in the future.
↑ comment by habryka (habryka4) · 2024-12-12T22:40:57.984Z · LW(p) · GW(p)
For the record, I wrote it late at night and ran the response through an LLM to improve readability for her benefit.
This is IMO generally considered bad form on LW. Please clearly mark whether an LLM was involved in writing a comment, if the final content is not something that genuinely reflects your own voice and writing style and you hold it to the same standard as your own writing. Like, it's OK to iterate on a paragraph or two with an LLM without marking that super prominently, but if you have a whole comment which clearly is a straightforward copy-paste from an LLM, that should get you downvoted (and banned if you do it repeatedly).
Replies from: Seth Herd↑ comment by Seth Herd · 2024-12-13T02:17:40.299Z · LW(p) · GW(p)
I assume you're in agreement that the reason for this is as nicely stated by cata in this thread: LW contributors are assumed to be a lot more insightful than an LLM so we don't want to guess whether the ideas came from an LLM. It's probably worth writing a brief statement on this unless you've added it to the FAQ since I last read it
Replies from: habryka4↑ comment by habryka (habryka4) · 2024-12-13T02:44:02.445Z · LW(p) · GW(p)
I think beyond insightfulness, there is also a "groundedness" component that is different. LLM written writing either lies about personal experience, or is completely absent of references to personal experience. That makes writing usually much less concrete and worse, or actively deceptive.
↑ comment by gwern · 2024-12-13T00:03:14.873Z · LW(p) · GW(p)
Why not post the before/after and let people see if it was indeed more readable?
Replies from: haotian-huang↑ comment by Haotian (haotian-huang) · 2024-12-13T05:16:26.957Z · LW(p) · GW(p)
Before:
Aspirin is a credible better candidate than Tamiflu. First off safety profile because you need to weigh up risks and benefits. Short term usage (what you're interested in for acute viral infections) is associated with GI upset, but only marginally more than placebo. There are of course other considerations that should be taken into account, but for the majority of users short term usage of aspirin is safe. Special note here about the association with aspirin and Reye's syndrome in children, the incidence of this is extremely low, but it is worth considering if your patient is under 16. In terms of efficacy against H5N1, it has demonstrated great success in vitro and in vivo acting via its NF-κB-inhibiting activity.
Methylene blue is also a credible better candidate than Tamiflu. Starting again with safety profile the gist is short term usage is generally safe unless you happen to be contraindicated. There are a few reasons and these briefly (NOT comprehensive do not take medicine based off this comment alone) (i) if you are at risk of serotonin syndrome (ii) known hypersensitivity (iii) severe renal impairment. If you are considering this medication I implore you to read at least the following data sheet. With that out of the way, how well does it work against H5N1? Efficacy-wise it has little-appreciated broad-spectrum antiviral properties. and has specifically demonstrated potent virucidal activity against H1N1 (H5N1 not yet tested to my knowledge). Interestingly, there is also a patent application for this indication. Taken together it could credibly be better.
For balance, Tamiflu is also associated with many of the same common adverse side effects including nausea, vomiting, diarrhoea or abdominal pain. Side effects typically occur along a distribution of individual susceptibility and should be tolerable for most for all three drugs. However, efficacy wise, aspirin and methylene blue are both credible candidates at outperforming the benefits of Tamiflu against H5N1.
After:
Aspirin
First, consider aspirin’s safety profile. For short-term use—relevant in acute infections—the risks are minimal for most individuals. GI upset, often cited as a concern, is only marginally higher than placebo (source: PMC3586117). The notable exception is its association with Reye's syndrome in children, but this is exceptionally rare. The risk-benefit calculus remains favorable for adults, particularly in acute scenarios.
Efficacy-wise, aspirin’s NF-κB-inhibiting activity shows promise as an antiviral. It has demonstrated robust effects against H5N1 in vitro and in vivo (source: Wiley). This pathway is a compelling mechanism of action that Tamiflu does not directly address, positioning aspirin as a strong candidate for consideration.
Methylene blue
Next, methylene blue is another plausible alternative. Its safety profile for short-term use is generally acceptable, barring contraindications such as risk of serotonin syndrome, hypersensitivity, or severe renal impairment (source: Medsafe). For those without contraindications, it represents a well-tolerated option.
Methylene blue's broad-spectrum antiviral properties are underappreciated. It has shown potent virucidal activity against H1N1 (source: PMC8275569), and while H5N1-specific data is lacking, the mechanism of action suggests potential efficacy. Notably, there’s already a patent application exploring its antiviral indications (source: WO2007086995A2). Given this, methylene blue is a viable contender worthy of further clinical study.
Comparing with Tamiflu
Tamiflu has well-documented side effects—nausea, vomiting, diarrhea, and abdominal pain—that overlap with the profiles of both aspirin and methylene blue (source: Medsafe Tamiflu). While tolerable for most, its mild benefits against H5N1 are unlikely to significantly outperform the mechanistic potential of aspirin or methylene blue (source: JHEOR).
Conclusion
In a pandemic preparedness scenario, Aspirin and Methylene blue deserve serious consideration alongside Tamiflu. They offer distinct mechanisms of action, robust preliminary data, and manageable safety profiles. If you're prioritizing interventions that maximize cost-effectiveness and theoretical efficacy, these "old drugs" might hold the key to new antiviral strategies. Rational prioritization of research and clinical trials could unlock their full potential.