5 ms·

Gradient Descending Through Brinespace

47 points by voisin 3y ago

light_hue_1 3y ago
This is a fascinating. A community I've never heard of before and I've never seen discussed in any sci-fi book. It's science cosplay! A community that has all the basic trappings of science. They collect data. They make graphs. They run "trials". They have hypotheses. They read papers. They publish "results". They start from a great place. And they keep all the fun parts of science. But they're missing all of the boring hard parts that make science work. The making sure your trials have controls. Designing trials carefully to avoid other interpretations. Running trial after trial to prove something. And the way they go about data analysis is 100 years out of date. They're essentially reading case studies. Trying to read into every single outlier and unpack them. There's a reason why that's not mostly forbidden these days. You just delude yourself into thinking that everything works! They also never have controls as far as I can tell, even obvious ones. If you adopt their methodology you'd prove that Pluto influences your diet. Keeping the fun things about science is a common problem with new and prospective grad students. It's what you really need to watch out for: you'll get a student who is extremely busy but woefully unproductive. The world is full of people who want to do fun things! Look at big problems and take some step in some direction. But then, there are very few people out there who are willing to spend the years it takes to cross all of the t's and dot all of the i's to really prove something in a scientifically useful way.
- mold_time 3y ago
  Hi! We think that the way science was done 100 years ago was in fact better and we are consciously emulating parts of the old-fashioned approaches that we like. :) What do you think of this: https://slimemoldtimemold.com/2022/01/11/reality-is-very-weird-and-you-need-to-be-prepared-for-that/ https://slimemoldtimemold.com/2022/01/11/reality-is-very-wei...
  - light_hue_1 3y ago
    I think you're romanticizing a past era and idea of science. Approaches aren't old fashioned for arbitrary reasons! They're old fashioned because they don't work. Science is pretty cutthroat. No one wants to leave anything on the table. If those methods worked to discover big ideas people would be pushing them. There are plenty of people out there who are what I'd call crash and burn scientists. The kinds of people who bet their entire careers on doing one big splashy thing. If it works out, great! If it doesn't, they're out of a job. If the old timey methods worked for finding such ideas people would be pumping them hard. But don't. And there's a reason why we hardly made any progress until about 100 years ago when we stopped doing science this way. Splitting subgroups isn't inherently bad! Heck, something like stratifying autism is one of those big problems people burn their entire careers on. But after trying to do this based on case studies for decades, we've learned it basically doesn't work as a method. Splits come out from mechanistic hypotheses and/or in the data at scale when you have controls. I do agree that the old-timey way of doing this is much more pleasant. But it's just not effective or what we would call science today. Even if you look at the case of Omegaven. It was the animal studies that saved them. Boring science with controls, etc. The main problem they had is that funding agencies are too conservative. Which is true! The vast majority of scientists would agree with that statement. We need to take more risks and funding agencies need to be ok with more failures. Sadly, funding agencies don't agree. I'm not writing this to be mean! I'm writing it because I see the impulse to follow this approach to science all the time in new grad students. I think you could be far more effective if you took a different much more scientific approach. Citizen science has a real chance of completely revolutionizing nutritional science because our studies are so incredibly bad. But the solution isn't to throw the science part out, it's to fix the part that we know is bad: small studies with bad controls and really poor reporting. We're a good app + a large community that's willing to play ball away from changing the world. Happy to chat more :)
    - Pannoniae 3y ago
      While it's true that conservatively performed science does lead to more consistent results (so not every correlation would be claimed as a brand new discovery), nowadays we have two basic problems. 1. Science has become so abstract that it's very easy to cheat with statistics and methodology. This is both due to the enormous complexity which makes it hard to verify, but also because there are few low-hanging fruits, and people are pressured to show a result. I even had that pressure in high school in some classes to show results from nothing! 2. There are many areas which are so advanced that the overwhelming majority of us, and even experts barely comprehend any of it, but simultaneously there's many things which we barely know anything about. Sadly or not sadly, many of these areas are simply neglected because universities and companies aren't interested in them. That doesn't mean that these areas are useless or impossible to research, they are just either niche or non-existent. Now, people in those areas obviously won't use the "new" scientific methods since they don't even know themselves what will happen or how it works, so it's not possible to theoretise it in a similar way, unless if they randomly make them up. Also, many (I would even say most) publications can safely be cut down by 50% without losing anything of value. I am saying this because you mentioned science is cutthroat - but it isn't cutthroat in coming up with new experiments, tools and making observations, it's mostly cutthroat in academia politics, falsifying/p-hacking data and writing eloquent bullshit in the form of shibboleths designed to be read by other peers in academia.
- alfiedotwtf 3y ago
  Are you saying Metaheuristics are is science cosplay?
- voisin 3y ago
  Have you considered that having a diversity of approaches might be of value, particularly when this approach (“citizen science”) has very low to no cost and could lead to a “real scientist” to think “hmmm… let’s do a real study on that”? I.e. perhaps low cost science should have a role leading the science peloton, cutting through the friction so that more resource intensive science can go where the spotlight shines? Just a thought.

47 points by voisin 3y ago

This is a fascinating. A community I've never heard of before and I've never seen discussed in any sci-fi book. It's science cosplay! A community that has all the basic trappings of science. They collect data. They make graphs. They run "trials". They have hypotheses. They read papers. They publish "results". They start from a great place. And they keep all the fun parts of science. But they're missing all of the boring hard parts that make science work. The making sure your trials have controls. Designing trials carefully to avoid other interpretations. Running trial after trial to prove something. And the way they go about data analysis is 100 years out of date. They're essentially reading case studies. Trying to read into every single outlier and unpack them. There's a reason why that's not mostly forbidden these days. You just delude yourself into thinking that everything works! They also never have controls as far as I can tell, even obvious ones. If you adopt their methodology you'd prove that Pluto influences your diet. Keeping the fun things about science is a common problem with new and prospective grad students. It's what you really need to watch out for: you'll get a student who is extremely busy but woefully unproductive. The world is full of people who want to do fun things! Look at big problems and take some step in some direction. But then, there are very few people out there who are willing to spend the years it takes to cross all of the t's and dot all of the i's to really prove something in a scientifically useful way.

mold_time 3y ago
Hi! We think that the way science was done 100 years ago was in fact better and we are consciously emulating parts of the old-fashioned approaches that we like. :) What do you think of this: https://slimemoldtimemold.com/2022/01/11/reality-is-very-weird-and-you-need-to-be-prepared-for-that/ https://slimemoldtimemold.com/2022/01/11/reality-is-very-wei...
- light_hue_1 3y ago
  I think you're romanticizing a past era and idea of science. Approaches aren't old fashioned for arbitrary reasons! They're old fashioned because they don't work. Science is pretty cutthroat. No one wants to leave anything on the table. If those methods worked to discover big ideas people would be pushing them. There are plenty of people out there who are what I'd call crash and burn scientists. The kinds of people who bet their entire careers on doing one big splashy thing. If it works out, great! If it doesn't, they're out of a job. If the old timey methods worked for finding such ideas people would be pumping them hard. But don't. And there's a reason why we hardly made any progress until about 100 years ago when we stopped doing science this way. Splitting subgroups isn't inherently bad! Heck, something like stratifying autism is one of those big problems people burn their entire careers on. But after trying to do this based on case studies for decades, we've learned it basically doesn't work as a method. Splits come out from mechanistic hypotheses and/or in the data at scale when you have controls. I do agree that the old-timey way of doing this is much more pleasant. But it's just not effective or what we would call science today. Even if you look at the case of Omegaven. It was the animal studies that saved them. Boring science with controls, etc. The main problem they had is that funding agencies are too conservative. Which is true! The vast majority of scientists would agree with that statement. We need to take more risks and funding agencies need to be ok with more failures. Sadly, funding agencies don't agree. I'm not writing this to be mean! I'm writing it because I see the impulse to follow this approach to science all the time in new grad students. I think you could be far more effective if you took a different much more scientific approach. Citizen science has a real chance of completely revolutionizing nutritional science because our studies are so incredibly bad. But the solution isn't to throw the science part out, it's to fix the part that we know is bad: small studies with bad controls and really poor reporting. We're a good app + a large community that's willing to play ball away from changing the world. Happy to chat more :)
  - Pannoniae 3y ago
    While it's true that conservatively performed science does lead to more consistent results (so not every correlation would be claimed as a brand new discovery), nowadays we have two basic problems. 1. Science has become so abstract that it's very easy to cheat with statistics and methodology. This is both due to the enormous complexity which makes it hard to verify, but also because there are few low-hanging fruits, and people are pressured to show a result. I even had that pressure in high school in some classes to show results from nothing! 2. There are many areas which are so advanced that the overwhelming majority of us, and even experts barely comprehend any of it, but simultaneously there's many things which we barely know anything about. Sadly or not sadly, many of these areas are simply neglected because universities and companies aren't interested in them. That doesn't mean that these areas are useless or impossible to research, they are just either niche or non-existent. Now, people in those areas obviously won't use the "new" scientific methods since they don't even know themselves what will happen or how it works, so it's not possible to theoretise it in a similar way, unless if they randomly make them up. Also, many (I would even say most) publications can safely be cut down by 50% without losing anything of value. I am saying this because you mentioned science is cutthroat - but it isn't cutthroat in coming up with new experiments, tools and making observations, it's mostly cutthroat in academia politics, falsifying/p-hacking data and writing eloquent bullshit in the form of shibboleths designed to be read by other peers in academia.
alfiedotwtf 3y ago
Are you saying Metaheuristics are is science cosplay?
voisin 3y ago
Have you considered that having a diversity of approaches might be of value, particularly when this approach (“citizen science”) has very low to no cost and could lead to a “real scientist” to think “hmmm… let’s do a real study on that”? I.e. perhaps low cost science should have a role leading the science peloton, cutting through the friction so that more resource intensive science can go where the spotlight shines? Just a thought.