• 0 Posts
  • 14 Comments
Joined 3 days ago
cake
Cake day: November 18th, 2026

help-circle

  • I have doubts about the unprofitability of the specialized models. Claude has already proven incredibly useful for programming. There are tangible results (see the bottom of this comment). The average person isn’t following the any of the tech niches at all. They simply want to believe that LLMs can only produce garbage code that falls apart.

    I agree that every day models like the casual chatbots aren’t likely to be profitable. Nobody’s going to want to pay expensive subscriptions to randomly chatter with a bot.

    The profit is going come in selling the specialized models. That’s why firms are rushing to scoop up all the professional talent who are willing to come work as AI model trainers. This is where the profit is. It’s actually better if they don’t create a single all purpose generative AI. This is not the end game.

    They can charge subscription fees by profession with multitudes of different models. They can divide up the models into Robertson screwdriver bot. Philips screwdriver bot. Flathead screwdriver bot. “Sorry, this model doesn’t include the torx head package. Click here to add torx model to your account!”.

    Anthropic specializes in LLMs for programming. There is certainly a future demand for this type of product. Whether it’s Anthropic or another company that ultimately comes out on top is anyone’s guess.

    Here are some examples of tangible results produced by Claude. Prior to LLMs these projects were impossible or would’ve taken years and years for some hobbyists to even begin making them. The latter two had failed attempts by people to do it without the aid of LLMs. Suddenly with Claude, this code got knocked out in a matter of weeks. These are working projects that people are actually using. Not the slop code that average people like to laugh at and ridicule.
    https://aioscilloscope.com/
    https://github.com/bve/OpenOSC
    https://github.com/kralonur/kws-x1-standalone-rs

    There’s certainly a demand for this type of AI. Somebody will find ways to make a profit off it. The technology is moving fast. People really want it to fail. That’s causing them to not perpetually reevalute objectively how things are changing.

    Day to day stock prices hardly tell much. It’s convenient that people like to say the stock market is meaningless when it’s going up. When it starts to crash, suddenly it’s meaningful.


  • StackOverflow had actual pedantic grumps. Reddit is all about gaming the votes. One tactic is to downvote neighboring posts so your own gains more traction.

    It could also be that reddit is used like a personalized feed. Not a community board. People downvote because they’re trying to tailor their feed. They don’t care about you or your question. They just want to see content that appeals to them.



  • What’s they’re overall point? I can be assed to listen or watch every hours long link or research paper source bombs that random commenters post.

    Surely they aren’t stating that AI is going away completely. AI predates LLMs and the industry will continue to exist with or without LLMs.

    I’m guessing they’re saying the current boom won’t last? Am I close? That’s what I’ve said in original comment too.

    Unless your point was to conflate the inevitable end of boom with AI going away completely. Use your words and explain. Don’t go dropping links and expect people know what you’re trying say.




  • Yep. Older generations of my family did piecework back in the day too. Initially I hadn’t thought about that. It came to me a few months into doing it. We’ve circled back to where we were a few generations ago.

    Piecework is about fixed rate, but this is essentially what it is because we’re measured by the rate of submissions. Some projects are hourly pay. Others paying per unit of production.

    Some companies are quite bad. They won’t pay for submissions that get a low grade. If it takes too long then the submission is paid lower or not at all.


  • I just got banned from one of these platforms. It was hell working that job. I really need money though.

    Everyone is constantly paranoid about getting banned. They require invasive logging to be installed. Everything is monitored. It’s micromanagment on steroids.

    One moment you’re looking at a picture of a thing, intensely scrutinizing it, hoping you’re working fast enough too. Suddenly the page auto redirects to a ban message. All your access is revoked. I spent a year fearing that every single day and it finally hit me. People wake up every morning to check and see if they’re not removed.

    My mind turns to mush grinding out data labelling/validation all day long. The pressure to meet rate and not mess up is intense. Your work is graded and you receive a score on each submission. It’s a lot of pressures.

    They are rolling out full time positions. So expect to see that in the coming year. I personally could not do 40 hours of this a week. This a consensus among workers too. There was chatter about roughly 4 to 6 hours a day of this kind of work at the most. Some did say they log the full amount of time we were alotted per week. That’s 10 or 12 hour days. Could you scrutinize photos for 12 hours a day? You have do it on an average of let’s say 2 or 3 minutes per photo. It usually require quick write up about it. It has to be perfect so as to not induce errors in the AI. You will receive a grade on each submission which is done by another worker like you who is under the same pressures.

    I’ve seen some of work people submit. Quite frankly the impoverished and less educated regions submit junk data. There’s no other way to say it. There’s more errors. The clarity of writing is worse too. Language models are pedantic about phrasing so fluency matters.

    They appear to be favoring workers from western regions lately. That’s only my gut feeling though. The pay can be quite high if you’re a professional in just about any field. Not only STEM. I personally did not see them pull tactics to reduce pay. Mine actually went up. I have heard the stories though. They pay third world countries a fraction of what they pay western countries for the same work. Then again, the quality kind of reflects that.

    Some who have done the job of reviewing work have expressed negative feelings about it. The job is to review other’s ability to label data. But they’re determining someone’s livelihood. Other’s don’t care. Some relish in the power they get from it.

    ETA: I forgot to mention. My computer would randomly lag while working. I assume it was whatever invasive surveillance their software was doing.




  • UnethicalGallon@lemmy.worldtoMemes@sopuli.xyz•Me neither
    link
    fedilink
    arrow-up
    24
    arrow-down
    1
    ·
    3 days ago

    Does anyone remember Y2K. There were entire sections of stores dedicated for Y2K preppers. Actual world events might be crazy lately, but people aren’t Y2K level of crazy yet.

    My uncle was stockpiling bottled water in his basement in preparation for computers never working again.

    Now that I think of it. We could probably do with no more computers ever right about now.