The_Grinch [he/him]

Have I truly become a monster?

  • 3 Posts
  • 216 Comments
Joined 3 years ago
cake
Cake day: January 4th, 2023

help-circle






  • Is there some technical reason why they can’t fix the AI hallucination problem by hooking into a traditional “knowledge engine”? Wolfram alpha has existed for a long time now. It lets you query a database of facts about the world, and even do math on those data. Why couldn’t AI just rephrase your question into a query for a beefed up version of that and then it would automatically show its work and use traditional computing which is actually good at math and logic instead of being a very confident “liar” which seems to be an unfixable problem with LLMs?

    I know AI will buck using say, python, to do a calculation, claiming it doesn’t need to because it “knows” it’s right, but what if it could only respond to a question through something like wolfram alpha?

    If they put just one of their billions into building out a hand curated knowledge engine it would be much more useful more quickly than their market competitors who are still trying to make an LLM do something technically impossible.




  • That’s because they’re engineering their desktop for first time users who look first, then click. Having things visually “tidy” without too much “clutter” or anything that might make them feel overwhelmed is what they’re looking for. Being predictable, consistent, or able to learn by muscle memory is less important. If you’re measuring success based primarily on increasing number of users, onboarding is by far the most important aspect of design.

    Seasoned users of a piece of software know exactly where the button/menu/tool they want is, and their needs are often directly contrary to a first time user’s needs. These users want the element they’re looking for to be accessible in as few actions and little thought as possible.

    The ideal software that you would use day to day is able to be approachable, but holds your hand while you become a seasoned user. Menubars were ideal for this. Every function is laid out for new users to look through. You have spacial memory for where each function is organized. On MacOS and a couple linux desktop environments functions with a keyboard command associated would have that command displayed beside them (and you can even set one if one doesn’t exist, or change one that does), gently assisting you to use the program more easily. Several desktops also offer searchable menubars which is just another layer of convenience. Big shiny buttons for common functions and a hamburger menu are simply a step backward from the traditional menu bar. You’re only a new user of a piece of software once.

    At best, GNOME, the party in control of GTK and design for a huge swath of software, refuse to play ball and cooperate with the rest of the linux/FLOSS desktop ecosystem. At worst they want to throw out all the literature about muscle memory, predictability, and familiarity in UI design and impose their frankly annoying Fisher-Price design on everyone else while calling you an out of touch elitist for resisting this.