Collection of papers and articles that I’ve spotted since my previous links post that seem interesting.

  • The history of Venture Research, the original Emergent Ventures (ht/ Patrick Collison for the pointer)
    • Restoring the conditions of the pre-1970s environment would, he said, require getting rid of boundaries, deadlines, milestones, peer review, priorities and any objectives beyond exploration and understanding.13 According to Braben, starting in the 1970s, increases in public funding for science brought demands for justification of that spending, which in turn encouraged scientists to produce ‘a virtually endless supply of short-term miracles’.12 This process, he says, has made fields more specialised, driven up the costs of discoveries and fuelled claims that the only discoveries left to be made are difficult and expensive – an idea he refutes.
    • Box 1 in the report has links to criticisms of the peer review process, and some possible alternatives.
  • The infamous Quillette N=22 study
    • Usually, small samples are suspicious and lead to results that do not replicate; here the effect size is so large (22:1) that it may be true (e.g. compare flipping a coin 22 times and getting 21 times head. Would you believe that the coin is fair?)
    • One may reply with one of Gelman's critique: That small sample sizes lead to magnitude (And sign) errors in the metrics of interests  often when the result is statistically significant. An effect size that huge would then be suspicious.
    • There's the other Gelman critique: The Garden of Forking Paths critique (That one can slice the data in many ways to get a significant conclusion). However, here the headline figure is just a simple ratio, and the difference is so large that it is unlikely to be due to the methods.
    • Instead, the only valid critique I see is in the sampling procedure, if the sample is biased then what follows is invalid.
    • For example, in that sample many of those were white nationalists or actual nazis; the claim would then be a more boring "twitter bans nazis more than leftists".
    • Ideally, we want to sample all conservative and leftist political commentators on twitter, and out of that, see how many got banned.
  • A new initiative for longevity research launches
  • Human capital interventions on entrepreneurial performance
    • human capital interventions are broadly defined to include business‐related education or training, coaching, entrepreneurship programs, formal education, and programs that foster informal or nonformal learning
    • The only intervention with a consistently positive effect on entrepreneurial earnings and firm profits is formal education, although the effects are small.
  • Boeing claims to be able to cut prices substantially with new aircraft design.
  • One more piece of evidence that dark matter may not actually exist after all, at least not in a way widely believed
  • In defense of cosmopolitanism
  • JP Morgan to launch their own internal cryptocurrency
    • Why? Official reason is to transfer money around between institutional accounts faster
  • Eight popular use cases for Blockchain, and why they do not work
    • It's oracles all the way down
    • Also, McKinsey report
  • The ant colony as a test for scientific theories of consciousness (ht / @SilverVVulpes)
    • The China brain experiment, redux.
  • Head of Theoretical Physics at CERN, Gian-Frencesco Giudice makes case for a larger collider, gets Sabine Hossenfelder'ed
  • Scott Alexander closes the Culture War thread
  • Top nutrition myths of 2019.
    • TLDR: Carbs, proteins, fat, beef, it's all okay. Just eat reasonably and you'll be fine.
  • Do you remember that study that said that blind auditions for orchestras increase the % of women that get admitted? Well, it doesn't seem to be true. Critical discussion on this twitter thread
  • OpenAI releases advanced general purpose text model GPT-2. Or rather, they release a limited version of it, citing concerns with malignant use.
    • Scott Alexander sees this as a step towards general intelligence
    • I do not, and I disagree that we are just _brute-force statistical pattern matcher which blends up the internet and gives you back a slightly unappetizing slurry of it when asked., _or that GPT works in a similar way as a brain. Though sure, there's a point there: everything is a remix
    • Compared to a brain, and as with every other impressive recent achievement in AI -and DL in particular-, the approaches require far more computational power, energy, and training sets.
    • As an example, GPT-2 can produce (after  a few trials) relatively good pieces of text, but still show that there is no underlying understanding behind, especially in the longer texts. (Some obvious inconsistencies).
    • What would impress me? A system that can generate, 100% of the time (Or, as often as a novice human writer can) coherent text, or text that could fool human validators in a Turing test. Bet alert: I bet this won't happen in 2019 or 2020, 100$, if you want to bet me, contact me.
  • I read David Reich's book on genetics, and I discovered among other things that people from Madagascar originally came from Malaysia; African migration into Madagascar happened later.
  • The evolution of healthcare spending in the US. There seems to be aslowdownin expenditure growth in recent times.
  • Reduction in costs of solar panels, brought to you by advances in engineering, not science.
  • Daniel Lemire comments on academia, and the motivations of researchers.

Comments from WordPress

  • Rational Feed – deluks917 2019-02-25T19:10:34Z

    […] Links 25 by Artir – Links post: Open AI’s language model, science funding, an n=22 study on whether Twitter is biased, nutrition and other topics. […]