3 comments

  • muzani 21 hours ago ago

    Quoting Paul Graham,

    John McPhee wrote that Bill Bradley's success as a basketball player was due partly to his extraordinary peripheral vision. "Perfect" eyesight means about 47 degrees of vertical peripheral vision. Bill Bradley had 70; he could see the basket when he was looking at the floor. Maybe great hackers have some similar inborn ability. (I cheat by using a very dense language, which shrinks the court.)

    https://paulgraham.com/gh.html

    It's doubly so with LLMs. You don't need as much clean code. Larger files. Make it easier to structure files for LLMs to navigate. Linting and tests that tells AI when it's making mistakes. AI doesn't always run the code, but something like Dart with hot reload, gives it much faster feedback.

  • lendacerda a day ago ago

    Yes you need good amounts of data for the LLM to learn the language. But after some point, the easyness of the language is more important. Python and Javascript (or TS) are kings

    Ideally, you dont wanna LLM to reason. You yourself should understand the problem and the better solution for it, and then the AI just executes on it

  • Bostonian a day ago ago

    I think you want a language where code that compiles is more likely to be correct code. Code that does not compile is only an annoyance, since it can't make its way into production.