There is no shortage of hype around AI coming for jobs, and while the U.S. labor market has begun to sputter, hard evidence of AI-related job losses is scarce.

There is no shortage of hype around AI coming for jobs, and while the U.S. labor market has begun to sputter, hard evidence of AI-related job losses is scarce.

Geoffrey Hinton’s message on a recent podcast about artificial intelligence was simple: “Train to be a plumber.”

Hinton, a Nobel Prize-winning computer scientist often called “the Godfather of AI,” said in June what people have now been saying for years: Jobs that include manual labor and expertise are the least vulnerable to modern technology than some other career paths, many of which have generally been considered more respected and more lucrative.

“I think plumbers are less at risk,” Hinton said. “Someone like a legal assistant, a paralegal, they’re not going to be needed for very long.”

  • MagicShel@lemmy.zipEnglish
    121·
    4 days ago

    Yes. I use AI every day at work.

    write a complex SQL query in a large database?

    This is not a great use case for AI generation because it combines long multiple select queries with having to know the DDL of all the relevant tables. It might get you an approximation that needs tweaking, but if you work with SQL every day I’m going to bet you’d do a better job at this in less time than it takes to have AI help.

    Things it’s bad at:

    • ingesting lots of code, and spitting back out a “fixed” version of that code.
    • Writing complex logic in one fell swoop.
    • Writing code that has a bunch of externalities to consider like exactly where you store user scripts on a given Linux installation or writing stuff that works locally AND in the CI/CD pipeline, etc.
    • Adding code to a codebase that contains a bunch of custom packages and non-standard designs

    Things it’s good at:

    • “Complete all the swagger annotations for http status codes 400, 401, 403 <…>”
    • “Review the standards provided in corp_api_design.md and validate conformance.”
    • “Describe the necessary endpoints, components and functionalities for a user service.” edit result and ask for review
      • “Given this architecture, stub out a controller, service, and any request and response objects.”
      • edits result and request review
      • “Add validations to <class>” repeat as necessary
      • “Given this controller stub, service stub and response object, write a controller unit test using JUnit5 and MockMvc.”
      • “Given <stub, etc.> implement <function>”
      • “Given <Entity/ies> and <ResponseDTO> create a mapper from entity to response. Use builder pattern and implement null checks.”

    Any of these results may need massaging by hand. The AI can’t do the whole job, especially at once. But it can write bite sized pieces, sometimes even mouthfuls, very quickly. It’s not instantaneous but it’s faster than doing it from scratch. For me.

    But I get the most value out of having my work instantly reviewed. I miss stuff. I typo stuff. Yesterday it caught in seconds a spot where I’d put a similarly named but wrong class and had struggled for 30 minutes to see it. It noticed the pattern established in similar code wasn’t followed once I passed the unclear IDE error and code.

    I know my job well but my execution is imperfect. AI excels at noticing those variations and imperfections from the rest of the code or industry standards.

    • jaykrown@lemmy.world
      72·
      4 days ago

      That’s fair, so your productivity has increased and error rate has lowered due to responsible usage of AI.