• 0 Posts
  • 22 Comments
Joined 11 months ago
Cake day: July 30th, 2023
  • computergeek125@lemmy.worldtoTechnology@lemmy.world•Are there any genuine benefits to AI?English
    1·
    5 months ago

    Oh I am in fact giving the giant auto complete function little credit. But just like any computer system, an AI can reflect the biases of it’s creators and dataset. Similarly, the computer can only give an answer to the question it has been asked.

    Dataset wise, we don’t know exactly what the bot was trained on, other than “a lot”. I would like to hope it’s creators acted in good judgement, but as creators/maintainers of the AI, there may be an inherent (even if unintentional) bias towards the creation and adoption of AI. Just like how some speech recognition models have issues with some dialects or image recognition has issues with some skin tones - both based on the datasets they ingested.

    The question itself invites at least some bias and only asks for benefits. I work in IT, and I see this situation all the time with the questions some people have in tickets: the question will be “how do I do x”, and while x is a perfectly reasonable thing for someone to want to do, it’s not really the final answer. As reasoning humans, we can also take the context of a question to provide additional details without blindly reciting information from the first few lmgtfy results.

    (Stop reading here if you don’t want a ramble)

    AI is growing yes and it’s getting better, but it’s still a very immature field. Many of its beneficial cases have serious drawbacks that mean it should NOT be “given full control of a starship”, so to speak.

    • Driverless cars still need very good markings on the road to stay in lane, but a human has better pattern matching to find lanes - even in a snow drift.
    • Research queries are especially affected, with chatbots hallucinating references that don’t exist despite being formatted correctly. To that specifically:
      • Two lawyers have been caught separately using chatbots for research and submitting their work without validating the answer. They were caught because they cited a case which supported their arguments but did not exist.
      • A chatbot trained to operate as a customer support representative invented a refund policy that did not exist. As decided by small claims court, the airline was forced to honor this policy
      • In an online forum while trying to determine if a piece of software had a specific functionality, I encountered a user who had copied the question into chatgpt and pasted the response. It was a command option that was exactly what I and the forum poster needed, but sadly did not exist. On further research, there was a bug report open for a few years to add this functionality that was not yet implemented
      • A coworker asked an LLM if a specific Windows powershell commands existed. It responded with documentation about a very nicely formatted command that was exactly what we needed, but alas did not exist. It had to be told that it was wrong four times before it gave us an answer that worked.

    While OP’s question is about the benefits, I think it’s also important to talk about the drawbacks at the same time. All that information could be inadvertently filtered out. Would you blindly trust the health of you child or significant other to a chatbot that may or may not be hallucinating? Would you want your boss to fire you because the computer determined your recorded task time to resolution was low? What about all those dozens of people you helped in side chats that don’t have tickets?

    There’s a great saying about not letting progress get in the way of perfection, meaning that we shouldn’t get too caught on getting the last 10-20% of completion. But with decision making that can affect peoples’ lives and livelihoods, we need to be damn sure the computer is going to make the right decision every time or not trust it to have full controls at all.

    As the future currently stands, we still need humans constantly auditing the decisions of our computers (both standard procedural and AI) for safely’s sake. All of those examples above could have been solved by a trained human gating the result. In the powershell case, my coworker was that person. If we’re trusting the computers with at much decision making as that Bing answer proposes, the AI models need to be MUCH better trained at how to do their jobs than they currently are. Am I saying we should stop using and researching AI? No, but not enough people currently understand that these tools have incredibly rough edges and the ability for a human to verify answers is absolutely critical.

    Lastly, are humans biased? Yes absolutely. You can probably see my own bias in the construction of this answer.

  • computergeek125@lemmy.worldtoProgrammer Humor@lemmy.ml•Never againEnglish
    1·
    5 months ago

    I get the statement you’re trying to make here - serving the name of a platform you dislike with the same reverence as he-who-must-not-be-named in Harry Potter (Voldemort) - but all you’ve done is obfuscate the search engine. Now if someone is skimming for information on the platform via search, you’ve hidden your comments and post from someone who might find your perspective useful. No one is going to try 15 ways of spelling a platform name (except maybe trying stackoverflow with and without spaces). Internet users are pretty lazy.

  • computergeek125@lemmy.worldtoTechnology@lemmy.world•How in the hell is this possible?English
    1·
    6 months ago

    LLMs have a a tendency to hallucinate: https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)

    As someone else stated, the AI can’t reason. It doesn’t understand what a unicorn is. It can’t think “a unicorn has a singular horn, so a non existent two-headed unicorn would have two horns”. Somewhere along the line it’ll probably mix in a deer or a moose that has two horns, because the number two matches the number of horns per head statistically.

    Last year, two lawyers in separate cases with different LLMs submitted hallucinated case citations. It would have been trivially simple for them to drop the case number into a proper legal search engine, but neither did. This is a similar issue: the LLM will also prioritize what you want to hear, so it does what it’s designed to do and generate text related to your question. Like the unicorn example, it has no reasoning to say “any legal research should be confirmed by making a call to an actual legal database to confirm citations” like a human would. It’s just scribbling words on the page that look like other similar words it knows. It can make case notes look real as heck because it has seen other case notes, but that’s all it’s doing. (please excuse the political news story, but it’s relevant)

    And it’s not limited to unicorns or case notes. I found this reddit post while researching a feature of a software package (Nextcloud) several months ago. In the post, OP is seeking an option to pause the desktop client from the command line. Someone responds with a ChatGPT answer, which is quite hallucinated. Not only does such an option not appear in the documentation, there’s an open bug report to the software devs to request that the feature be added. Two things easy for a reasoning human to do, but the AI is just responding with what you want to hear - documentation.

    I’ve also seen ChatGPT tell my friend to use power shell commands that don’t exist, and he has to tell the model twice to generate something new because it kept coming to the same conclusion.