bschne

Motivation is funny like that. I've struggled to work through a paper I didn't even want to understand but had to write an assignment about on an A4 eInk tabled or using the perfect note taking setup, but I've also read a full book and 10k+ words blogposts on my phone or filled in a PDF crossword by placing single-letter text boxes in preview on my mac.

I've seen https://lu.ma used thought it's built more around one-offs than recurring meetups of the same group. Also, and I think it's closer to meetup, https://app.getriver.io/ -- though I don't think it has too much traction yet.

was wondering the same thing!

yeah they missed the tone a bit there, tripped over it too

For a semi-related case study of entrepreneurial tool-building in science, I also liked this: http://milan.cvitkovic.net/writing/neuropixels/

I couldn't help but think about mechanistic interpretability research on large neural models reading this — I guess this is what happens when humans do something similar, adding and removing tweaks here and there to better fit this or that case, over a long period of time.

Reading about this — I'm sorry to hear about your loss! — makes me wish we understood the dynamics of things like this much better than we do. Some stuff seems pretty clearly "not good", but sometimes it's really hard to reason about what is bad as long as you're exposed and leaves relatively little lasting damage, or what can have bad effects with highly delayed onset, and how the mechanism of "setting in motion the delayed-onset damage" works, exactly — is it the substance that stays in your body and continues to do damage, does it throw some process off-balance that then continues to harm you even after the substance is gone, etc. If anyone has any pointers to sources that cover our understanding of this better, I'd be very interested to read more about it.

I am serious! In an ideal world, Twitter/X would have more legitimate advertisers to keep the platform running, and also would have adequate content moderation that quickly got rid of the scams, through some combination of tech and human oversight. They are, and I am not debating that, doing pretty badly at that currently -- I get a few of the scam ads daily, ranging from misleading freemium games (sketchy) to sites faking local news publications and public figures to advertise financial scams (downright criminal fraud wherever you decide to draw that line).

But, looked at separately, allowing users to put notes under paid-for (!) ads pointing out issues with them is pretty remarkable. Tech gets a lot of flak for bowing to advertisers in the name of $$$, but here it is the users who have some power to set misleading ads straight, with zero influence by the advertiser in question. That is laudable, even if in an ideal world this would be unnecessary, or at least less load-bearing in terms of preventing fraud!

Ironically Twitter/X, for all its recent criticism, comes out ahead on this stuff: at least there they get community notes saying it’s a scam fairly quickly

Nitpick but it was even harder, they wanted to keep the PLCs intact, report back normal operation, and change the control output specifically to destroy/damage uranium centrifuges

I love this take. Reminds me of how the Mechanical Turk apparently indirectly inspired someone to build a weaving machine b/c "how hard could it be if machines can play chess" -- https://x.com/gordonbrander/status/1385245747071787008?s=20

I'm unfortunately not familiar with that, but maybe someone else will see the connection

It simply says that if you want to evaluate all paths from Cambridge to Palo Alto that go through the airport, you only need to consider those that use the quickest way of getting from Cambridge to the airport, because all others can't be faster than that. This is different from getting stuck in a local minimum, it's merely a dynamic programming approach to reducing the computational burden of finding the global one.

See my sibling comment, can recommend this: https://xcelab.net/rm/statistical-rethinking/

I am partway through McElreath's "Statistical Rethinking" and I fully agree with this.

> (…) the decision was made to go with a design that, in theory, would give users a 50/50 chance of plugging it in correctly

I love how they write „in theory“

"Wealth is only produced when people work." does not imply every kind of work creates wealth at the same rate. I think the point was merely that creating money out of thin air without anything corresponding changing in the real world does not cause there to be more of what people actually want there to be more of.

> Scientific consensus is built up around a body work

Yes, and that process is supposed to include judging the merit of the individual studies that make up the body of work. A hundred shoddy papers don't make a convincing body of work all of a sudden, they make a pile of inconclusive trash.

I'm not disagreeing with the consensus you point at, I wouldn't know, but you can't use "consensus arises from a body of work" to argue inclusion of anything in that body of work.

I read this extension of the concept a while back that dives a bit more into what's actually missing in the "cargo cult" approach, and how to transition to actual productivity — https://metarationality.com/upgrade-your-cargo-cult#upgradin...

I think GP had a different kind of therapy client in mind, the one who's more looking to get to better know themselves or get help figuring out their life and their struggles, but far from having their mental issues prevent them from functioning in life. In these cases, community may really help, but it may also not — communities of close people are often very good at plastering over someone's biggest and most glaring issues by either validating them or just downright avoiding talking about it. I think the benefit of a hired professional in those cases is that they can look at things honestly without the burden of social dynamics and longstanding relationships.

> It's similarly unclear why the material seems to always float at an angle. It seems to track the field lines in some way.

I've seen some people talking about the LK-99 samples only having these properties locally due to internal structure or something like that, could that not explain the angle if one part of the sample had more "pockets" of material with these properties than another part?

Which was inspired by this, "Could a Biologist Fix a Radio?"

https://www.cell.com/cancer-cell/pdf/S1535-6108(02)00133-2.p...

My perspective on this is more like "the locals will work x amount of time to get by either way, and you are offering them to work x amount of time and get a better living, or maintain their standard of living at lower x." (replace "time" with "time x enjoyment lost from missed leisure time", i.e. opportunity cost of time, if you will). You win because you get the work done cheaper or get to get more work done than you would "back home", the worker wins because they have a better alternative they didn't have before.

This hinges on the fact that I assume that if you take a job for OpenAI over, say, working at the local market, or your local family business, or doing subsistence farming, ... (I don't know the employment history of Kenyans doing this) you do so because you consider the job at OpenAI, in your own judgment, the better alternative. This assumes some amount of agency in the decision, yes — I would not argue the same if you e.g. forced prisoners to do the work for below minimum wage or whatever and then went and said "hey at least they get two dollars a day more than zero" (I exaggerate). Also, again, this assumes you don't mislead them about the mental toll and then hold them "captive" with sketchy contracts or whatever.

It also seems to me that you have an underlying assumption that companies in the global north could collectively decide to outsource as much labor as they are currently outsourcing, at wage levels matching their "home countries", which I don't think is economically true.

As for your analogy, I think I see the equivalence you draw, but I'm not sure I agree with it. For me the most natural read would be that the "payment" here is the workers' time, the ring you throw is the pay for the job. In that case, the payment (time spent doing some kind of work) happens either way (weird world where you always have to pay for getting saved by some strange law of the universe, I guess), and you're just offering a better ring, or the same ring for less time.

You would not take this option because you prefer living in the U.S. (or wherever) and being employed there, but that's not the relevant counterfactual for the people doing this work, and it seems like it's preferable to other options they actually have, else it would hardly attract workers.

Don't get me wrong, companies have an obligation to guarantee some level of workplace safety and healthy conditions etc.! But arguing that it's exploitation to buy labour where it is more affordable is plain silly.

If semantic HTML is important for accessibility and for software to be better able to parse information out of it, and AI solves the latter, semantic HTML is now less important because some of the use cases that needed it previously no longer need it. If you take "less important" as a moral/value statement instead of in terms of total utility provided, and assume that AI will have zero accessibility benefits, it will merely be as important as today, which is still at odds with the assertion of the original article that it would become more important. N.B. this seems doubtful, given how e.g. you can now past a bunch of code into an LLM and ask it questions quite naturally -- something I can easily see adapted to e.g. better navigating apps using only voice and screenreaders.

> Problem #1: What about all the other features?

Is this necessarily a problem in the argument? If we consider e.g. color, and construct a blurry image that has a color distribution very similar to a real face, but doesn't fire the edge detection in the same way, we can still recognize it as a face. This could just mean that a sufficiently close match to other examples of the same class on any one of the strong "dimensions" of the image our brain processes is sufficient to make us recognize it, no? The author makes it sound like if edge detection is what makes us recognize line drawings, this implies that our brain discards every other feature in attempts to recognize visual input, but I don't think that's a sound conclusion.

I recall reading about this in Logic mag a while back: https://logicmag.io/china/bullet-time/

adi shamiR (I'm sorry)

> I get 3 reviewers and 3 completely different reasons to reject, often nonsensical

Could one problem with these kinds of reviews be that the reviewers have absolutely zero skin in the game? Like, you can write complete nonsense and nobody will blame you personally for it, you have no stake in the outcome of the submission/paper what so ever, and you're much more incentivized to come up with reasons to reject than to say "good enough". I realize the latter part can be good in theory, as it sets a high bar, but it also often feels like this goes awry when not kept in check somehow.

> (...) make sure to refence the Big boys you were close to but definitely not stealing from

Much as giving credit is important, this is unfortunately also how you end up with intro sections littered with references to the same ten papers everyone's read already just to back up some extremely vague/general statement.