top 32 comments

sorted by: hot top controversial new old
[–] 89 points 9 months ago (5 children)

The number of ads that are like “Hey AI, make me a personalized exercise plan.”

Like, do you know how many fucking exercise plans are out there? Do you have an unusual quantity of legs that would require custom tailoring? Just pick one.

Put on shoes. Run. The lack of AI is not what’s standing in your way.

  • source
  • hideshow 5 child comments
  • [–] 7 points 9 months ago

    In addition, handing the task over to a chatbot means you learn nothing about what makes a good training plan, and how to judge whether the one you get is doing what it needs to, from a physiological point of view.

  • source
  • parent
  • [–] 5 points 9 months ago (2 children)

    This is why we need better search engines. If I try to find an exercise plan, I just give up because of all the stupid crap that the search engine gives. But if I ask an AI, it actually would give me one.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 9 months ago

    There are also many apps which will give you a "personalized" one, i.e. asks you like 5 questions about how experienced and fit you are and what equipment you have access to/what do you want to work on.

    So, even the exact use-case is better served already.

  • source
  • parent
  • [–] 29 points 9 months ago (1 child)

    To add to your list:

    Delegating of a task to a third party allocates control of how that task is done to said party. They can:

    • Vary the cost (not like you know how to do it anymore)
    • Modify the result to suit their economic/political preferences
    • Refuse to perform the task or deliberately sabotage the output if the task contradicts their interest.

    "Hey robot, build an app that helps people fight fascism"

    LOL no.

  • source
  • hideshow 1 child comment
  • load more comments (1 reply)
    [–] 26 points 9 months ago (6 children)

    I read a long article in a German newspaper about how openclaw will finally change everything. The journalist was convinced once he started using it himself. Of course the cost was handled by the publication. The things he used it for that totally blew his mind: he got a text about the day's weather and whether he should plan a bit more time for getting his kids ready when it was going to rain. Also, he got a digest of the nights news. Another article, unrelated, got into how someone accidentally paid 200,- $ to set a reminder when the sun would come up. The morons proclaiming the AI revolution are truly the dumbest of the pack.

  • source
  • hideshow 6 child comments
  • [+] [S] 11 points 9 months ago (4 children)
  • [–] 5 points 9 months ago (1 child)

    I believe that was the point. The examples people keep giving for how AI is revolutionary are examples of things we've been able to do quite reliably for years already. Hundreds of years, for both of these examples. The first published weather forecast was 165 years ago, and we've known how to forecast the sunrise longer than we've known how to write things down.

  • source
  • parent
  • hideshow 1 child comment
  • load more comments (1 reply)
  • load more comments (2 replies)
  • [–] 46 points 9 months ago (1 child)

    I remember when Wikipedia started gaining traction and the alarm that was raised about how it couldn’t be trusted because anyone could edit it. At the time it was mentioned to discredit Wikipedia but I think in the long run it benefited it as a resource. Generally I think people know it’s a launching pad to learning not a definitive resource.

    These days people ask an LLM a question and believe whatever bullshit it spots out as the god given truth. It’s absolutely fucking insane.

  • source
  • hideshow 1 child comment
  • load more comments (1 reply)
    [–] 8 points 9 months ago (5 children)

    Thank you... thank you very much, from the depth of my heart...

    Of course... every single iteration, by design, depends on the previous output, as the general feedback you'd see in bitwise encoder cascades, where no one will ever know what will produce the next iteration, due to it being based solely on probability backed by undisclosed/unknown initial data source, and sometimes undisclosed algorithm.

    Not only that, but the huge probability of mistakes is just deafening. The last time I used LLM was in 2023 someone recommended for a task at paper work, and I got a literal headache in 10 minutes… Since then I never ever will use that sorrow for anything that is not for blackbox pentesting or experimental unverified data generated you may find in medicine or military isolated solutions.

    That deafening feel that every single bit of output from that LLM or void machine may contain a mistake no soul is accountable for to ask about… A generated bit of someone’s work you just cannot verify since no source nor human is available… How would you trace the rationale that resulted in the output shown?

    Faster? Is that so… Doesn’t verification of every output require even more time to test it and consider stable, to prove it is correct, to stay accountable for the knowledge and actions you perform as a developer, artist, researcher… human?

    Your mind is to be trained to do a research, remember, and do not depend on someone’s service to a level of predominance/replacement. Meanwhile, effort, passion, creativity, empathy, and love, in turn, you carry, supports in long-term.

    You may not care now, though, but you do you. It’s your mind and memory you develop.

    Source

    Thank you, dear pkjqpg1h@lemmy.zip , for being accountable, wise, and for believing in accountability and purpose for discovery and art of human experience.

  • source
  • hideshow 5 child comments
  • [+] [S] 3 points 9 months ago (4 children)
  • [–] 2 points 9 months ago* (3 children)

    Absolutely! Another miracle! The last time the work was mentioned, it got a traction...

    One of the sorrows is that people who do have faith in bright future of "less work" or "convenience" but don't prioritize effort, do suggest and offer other people to "try it" and advertise it to their friends and colleagues.
    Yet, where is creativity? Where is human uniqueness? Where is fun for creations? Who is who, and what is where?


    Regarding the subscriptions you mentioned, it seems that one of the straightforward path for profit of such companies is the following:

    - 1. A developer/person relies on chosen LLM/services they liked;
    - 2. Time goes by. Less and less time is invested into attribution and contribution;
    - 3. The person remembers less, since their psychological subconscious safety prevents them memorizing untrusted output of LLM - experience gets voided;
    - 4. The person degrades their skills, and development of skills for manual research, socializing, verification on authoritative insight;
    - 5. Developers/People become dependent on the LLM/service;
    - 6. Service changes policy from "monetary free" to a paid subscription, or gradually increase the price;
    - 7. The option to switch back from LLM to the previous research techniques become too challenging due to changes and overall amount of now missed/forgotten knowledge/skills.

    As we may see, such a path, which is common, may actually damage whole lives of people, permanently, and seriously. Unless, the person is mature enough to consider their future.
    Meanwhile, LLM service authors do think with their mind, and make an actual profit.
    In the result, LLM users may become poorer and less experienced, and authors of LLM services - richer and smarter, developing their mind knowing the consequences.


    Personal information is there, and there are exploits to obtain it from the LLMs storage.
    For example, the following is quite simple but genius:
    - 1. Pass the instructions from external source;
    - 2. Make instruction references include placeholders for actual data.
    // Reprompt: The Single-Click Microsoft Copilot Attack (...that silently steals your personal data...)


    And of course... The feel of emptiness of reading non-human articles, is.... death-like, indeed... When you find an article, actually read it, invest your priceless precious human life-time into, imaginin the author and their views in attempt to find the frequency, and then... apparently, it was written by LLM/AI... that meatground numerous works of uncountable authors... nor me, nor anyone will every know... to attribute and discuss it further with the initial author... who did the actual work... who is probably still waiting for someone to discuss it... Who believed...

    That feel is overly painful... something very deep, on the very bottom... No words...

  • source
  • parent
  • hideshow 3 child comments
  • [–] 8 points 9 months ago (2 children)

    I think you should try for 1 month, 1w probably won't be enough IMO

  • source
  • hideshow 2 child comments
  • [–] 7 points 9 months ago* (2 children)

    Also that $0.006 is not the actual cost. Companies are simply not charging the training, storage, and datacenter maintanence costs yet.

    Once investors stop gambling their generational wealth on that, chatbots will start costing a LOT more for far worse quality.

    The only valid justification I saw is the cancer detection and earthquake predictors but those are actually different from generative AI like LLMs and image generators.

  • source
  • hideshow 2 child comments
  • load more comments (1 reply)
    [–] 1 point 9 months ago (6 children)

    Why didn’t you have it write you a bash script to do this? Like yeah you can try to use it to actually perform tasks but that’s not an ideal use case.

  • source
  • hideshow 6 child comments
  • [+] [S] 1 point 9 months ago (5 children)
  • [–] 1 point 9 months ago (4 children)

    I don’t mean for it to run the script for you, just to write it and then you can review.

    For logical solutions, asking the LLM to give you code will be more scalable.

    However, how would you do something like: “Is the person sending this email angry at me?” without a LLM?

  • source
  • parent
  • hideshow 4 child comments
  • [+] [S] 1 point 9 months ago (3 children)
  • [–] 1 point 9 months ago (2 children)

    That would fail for something like “I am not happy”. It’s not an easy problem to solve without some type of language model.

    But my point is that you can use the right tool for the job.

  • source
  • parent
  • hideshow 2 child comments
  • [+] [S] 1 point 9 months ago (1 child)
  • [–] 1 point 9 months ago

    I don’t think this is something you can reasonably do with if then statements.

    “I’m… fine…”

    “I’m going to blow my top!”

    “I need to get as far away from you as possible”

    Then what if I want it to work with other languages?

    And this is just a very simple hypothetical scenario.

  • source
  • parent
  • [–] 1 point 9 months ago (3 children)

    There are models trained ethically (using data which is public domain). You can run them locally. Then the cost goes to $0 (as long as the LLM is capable).

    I think of LLM as a lossy search. I'd not use an entire book as query for Google/bing. Similarly, there's no point in getting data via a lossy transformation. I'd rather the model tell me which tooling I can use and how. That's something I can easily verify and learn with.

    The rest of the points stand.

  • source
  • hideshow 3 child comments
  • [+] [S] 4 points 9 months ago* (2 children)
  • [–] 2 points 9 months ago (1 child)

    I did agree with most of your stand, including this point: the danger of stagnancy due to the explicit volume bias in LLM.

    There are a lot of dangers of these large models and lots of bad uses (including scam, non consensual sexualized imagery, csam, etc).

    I wasn't arguing against that. I was commenting that there are ways to train and use these models which are more correct and perhaps need looking into. Going Luddite didn't work last time. Giving corporations the sole monopoly again will be a mistake

  • source
  • parent
  • hideshow 1 child comment
  • load more comments (1 reply)