Marginalia - 2025-09-08

Reading List

• The Actuary's Final Word on Algorithmic Decision Making
• Don't Build an RL Environment Startup
• Notes on Roadtrips
• Craft Is the Antidote to Slop
• The Training Imperative
• A Revolution in Thought? - Dr Ian McGilchrist
• Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis
• Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task
• Claude Code: Behind-the-scenes of the master agent loop
• [A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox]((https://www.lesswrong.com/posts/vidXh2DJtnqH5ysrZ/a-blog-post-is-a-very-long-and-complex-search-query-to-find)
• Language Self-Play For Data-Free Training
• MoonshotAI/checkpoint-engine
• Fei-Fei Li: World Models and the Multiverse
• Performative Thinking? The Brittle Correlation Between CoT Length and Problem Complexity
• How Information Flows Through a Transformer (X Thread)
• Fixing the Biggest Problem With Mechanical Keyboards - this is not really about mechanical keyboards, it's an exploration of craft
• The courage to speak
• Color AI - Generate harmonious color palettes using AI
• Jonathan Blow on resisting success

  • As you become successful in your field (or wherever), and further internalize the habits that are necessary to be successful, it's clear that many of these things are easy to do, it's just that people don't want to do them. In other words ... it's obvious that many people don't want to be successful, and if they were to introspect deeply, they would see this clearly. In fact what they want is to be somewhere comfortable in the middle of the herd, not having to do too much work. Most people want to be comfortable, not 'successful' in a way that requires ambition. But many people are brainwashed enough by the rhetoric of success that they don't realize it's not what they want. There's also something I haven't figured out yet. Every time I give advice, I get a number of responses from people with self-defeating attitudes, explaining how this advice can't possibly apply to them because blah blah blah. These people build up belief structures that are obviously intended to keep them mired in their current situation, smelling of low self-esteem and defeatism. "Obviously" it's better not to be stuck in these belief structures, yet people will defend them vigorously, and in some cases fiercely. I don't yet fully understand why, except maybe that if someone believes there is a solution to their problem, then it must be their fault that they haven't solved it, and/or that there will be a clear failure that is their fault if they attempt to solve it.

Rough Thoughts in Progress

• Per Chris Dixon - your payoff is a $Max(attempts)$ not $Avg(attempts)$ function. There is basically no penalty for rejection in any endeavour - jobs, dates, VC intros, startups.
• Could there exist some kind of what3words for cryptographically verifying a message to mitigate deepfaked messages.
  • We obviously exist in an era where bad actors could feasibly spoof the face and voice of anyone close to you. Pair that with an emotionally charged message and the potential for abuse is massive.
  • w3w-like in the sense of simplicity and ease-of-use, not necessarily that the form factor is 3 words BUT that could logical too.
  • You could imagine some kind of multi-factor auth type system whereby the recipient can verify the authenticity of the message.
  • There are 2 distinct threat models to analyse:
    • 1. Attacker controls the number (but not the device).
    • 2. Attack controls the device (implication: the attacker holds the private key)