• 0 Posts
  • 70 Comments
Joined 1 年前
cake
Cake day: 2024年10月9日

help-circle

  • tl;dr: AI! Agents! AI! Agents! AI! Agents! AI…

    Just one thing that caught my attention:

    AI code review helps developers. We … found that 72.6% of developers who use Copilot code review said it improved their effectiveness.

    Only 72.6%? So why the heck are the other almost 30% of devs using it? For funsies? They don’t say.

    You’d think due to self selection effects most people who wouldn’t find using Copilot effective wouldn’t use it.

    The only way that number makes sense to me is if people were force to use Copilot and… no, wait, that checks out.









  • I have some thoughts about this goober (Simon Willison) that I need to get out of my head:

    First the positives:

    • I think he’s actually an experienced software engineer.
    • I think he care to check and test the LLM output.

    But, by his own admission:

    • He uses LLM for tasks he knows well (So easier to check and little negative impact on learning)
    • He works mostly on hobby projects (so no obligation to actually maintain the stuff)
    • He can choose to not use new libraries (which in a professional setting is not always a luxury you can afford)

    Tl;dr: an experienced dev who uses clankers to churn out tons of technically functional hobby software and thinks this gives him right to speak for all software engineers.





  • If we’re talking about the general West, then there new nuclear is probably fucked. Rest of the world still builds for reasonable costs. Not nuclear bro amounts, but still.

    I think we could see a future where nuclear makes 5-10% of the world’s electricity, which would technically make it a niche source of power, but it would also be a massive increase from today.