stupid ideas are good ideas: Nonlinear Function
Created: March 08, 2020
Modified: January 24, 2022

stupid ideas are good ideas

This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.
  • Revolutionary ideas must live in the blind spots of the current intellectual conversation; otherwise people would already be using them (the discourse is wrong). So they'll often look weird or stupid to 'elites'.
  • New ideas will not have been developed as deeply as existing ideas, so they'll seem less sophisticated.
  • For example: all of classical statistics is about carefully dotting i's and crossing t's to ensure that the conclusions we draw from noisy data are true. Statisticians prove properties of estimators, derive error bounds and confidence intervals.
    • Meanwhile, machine learning ignored all of those niceties and just fit parameterized functions to data with a predictive loss. As a field we realized that we don't understand the hard problems well enough to solve them with rock-solid theoretical guarantees, so if we want to make progress we can't let ourselves be too bound by rigor.
  • If SuccessfulFriend succeeds in disrupting computer systems, it will be because he recognized the value of a mindset and set of methods (Frank's work on differential dataflow) that was too different from the main CS systems community for them to care.
  • Being young is an advantage in recognizing grand new ideas, because
    • you're not trapped by being part of the existing elite conversation, and
    • while learning a field from fundamentals you have lots of time (and motivation) to synthesize your own view of how those fundamentals fit together, and
    • no one expects you to be performing at the top of your field (or really at all) yet, so you have room to explore crazy things. your bar for your own accomplishments is lower. I used to be able to get excited about a project just because it would help me learn some obscure thing. Now I can't get excited about a project unless I can see how it'll change my life. (I should change this though)
    • you're often in a school environment surrounded by other young people with the same advantages, and conspiracy is a thing for a reason.
    • Young people don't have learned helplessness around tackling big problems.
  • Related reading:
    • there must be a lot of this in the startup literature? pg's essays, zero to one, books on disruption?
  • What are my 'stupid ideas'? (these are sort of 'things I believe that no one else believes'---but Thiel's formulation implies a crisp and potentially actionable articulation, while 'stupid ideas' are allowed to be vague and unformed).
    • the connectionist thesis: soft, reconfigurable, floaty computation is going to be fundamentally more relevant to AI than precise algorithmic computation. (even if the algorithmic or symbol-manipulating perspective is highly useful for practical problem solving and for building domain-specific superhuman agents).
      • this is barely a 'stupid idea' because it's still one of the most dominant theses in current ML research---even if the rigor police are pushing back.
  • life conclusion: don't be afraid of exploring 'stupid' ideas, or ideas that I can't quite justify or articulate to elites. Neurodiversity can be a virtue. That said, remember that most stupid ideas are stupid ideas.