• 0 Posts
  • 18 Comments
Joined 2 years ago
cake
Cake day: June 13th, 2023

help-circle



  • According to the paper, they tested ten different split-and-merge scenarios and this one was the most likely. But they give some important caveats, including:

    • They assume that the smaller group had a more-or-less constant population size—if it fluctuated significantly, some of their other predictions on the dating of the split and merge might be off.

    • They can’t rule out more complicated scenarios, like three or more splits and merges (but they can rule out the simpler scenario of no splits).

    They do say that they tested their model on a number of other species (including chimps, bats, and dolphins), and got results consistent with those species’ known evolutionary histories.



  • They compared the entire genomes of 26 different modern human populations, and modeled their history to account for the patterns in the modern genomes.

    For example, suppose a particular gene has two distinct groups within the modern genomes, with each group showing similar mutations within the group that are different from the mutations in the other group. You can infer that the two groups represent a split into two populations that later recombined, and you can infer the time of the split and the relative population sizes of the two groups from the number of mutations in each group.

    Do that for the entire genome and you can make finer-grained inferences, like determining which genes experienced positive or negative selection pressure.






  • In previous work, Fedorenko and her students have found that computer programming languages, such as Python — another type of invented language — do not activate the brain network that is used to process natural language. Instead, people who read computer code rely on the so-called multiple demand network, a brain system that is often recruited for difficult cognitive tasks.

    I’m curious if there’s some overlap between conlangs and programming languages, on the region level if not the network level. IIRC, the multiple demand network is a bit ill-defined and every component doesn’t necessarily activate for every task; and Fedorenko et al have their own idiosyncratic definition of the language network that omits anything that might also have other functions (including canonical regions like Broca’s and Wernike’s areas).


  • In theory, authoritarianism is the fastest way to transform a society from one form to another—so it’s rational that regimes that take power based on the promise of rapid social transformation will be drawn to authoritarianism.

    But it’s also rational for institutions to try to preserve themselves—which for these authoritarian regimes means preserving the conditions that led to the belief in their necessity, instead of delivering on the promise of transformation that would lead to their dissolution.



  • If they tell law enforcement they can’t produce an unencrypted copy and it’s later proven that they could, the potential penalty would likely be more severe than anything they could have gained by using the data themselves. And any employee (or third party they tried to sell the data to) could rat them out—so they’d have to keep the information within a circle too small to make use of it at scale. And even if it never leaked, hackers would eventually find and exploit the backdoor, exposing its existence. And in either case they’d also have to face lawsuits from shareholders (rightly) complaining that they were never warned of the legal risk.