Microsoft has published an ad promoting Copilot on Windows 11, but it contains a hilarious error that actually proves how useless the AI is.

you are viewing a single comment's thread
view the rest of the comments
[+] -16 points 1 year ago (11 children)

Aura quickly manually sets the scale to 200%, which solves the problem but isn't what Copilot actually suggested…

Isn't that what almost anyone would do?

I've seen some pretty silly AI blunders before, but this one seems rather harmless. You're still going to end up at the setting you need to change to solve the problem, which to me falls squarely in the "close enough" bin.

  • source
  • parent
  • hideshow 11 child comments
  • [–] 41 points 1 year ago (4 children)

    Copilot didn't suggest 150% in the context of a set of instructions. He specifically asked Copilot which percentage to set. That was the only task which Copilot was being asked to solve at that point in time.

    You might also say that this is a simple task, so nobody would actually ask Copilot what percentage to use but isn't that the point? This task, which Copilot failed on, is a simple task.

  • source
  • parent
  • hideshow 4 child comments
  • [–] -1 points 1 year ago (1 child)

    He specifically asked Copilot which percentage to set.

    Ah, I see. That's not at all clear from what the article says:

    In the video, the influencer activates Copilot with the "Hey Copilot" voice activation and then asks it how they can increase the font size via Windows Settings.

    Perhaps the original video has what you're talking about.

  • source
  • parent
  • hideshow 1 child comment
  • Yes, it does. It's only a short video but, anyway:

    He has already opened the drop-down, where 150% is currently selected. He asks "what percentage should I click?" and Copilot replies "let's start with 150%, which is the recommended size". He hesitates slightly, then says "let's...", clicks 200% and then says "boom!".

  • source
  • parent
  • [–] 9 points 1 year ago (5 children)

    the problem is two fold and this applies to all LLMs.

    1. according to the article the person asked HOW to increase the font size. Copilot then tells Aura to go increase the font size so it's not exactly what the person asked.

    2. like all LLMs it MUST provide a solution. it MUST do something. Copilot asks Aura to increase the font size to 150%. it's already at 150%. but instead of going back to the user and saying "the font size is already set to 150%, would you like to increase it more?" it just goes ahead and bumps it up to 200 because it has to. it has no choice, it must provide some kind of solution. LLMs like GPT5, Claude, etc will do similar crap which ultimately shows that recent models are all collectively garbage. they now must ALL provide some sort of solution even if the majority of the time said solutions are actually hallucinations. LLMs aren't allowed to say "I don't know" or "it's already done" or whatever.

    So this video/article just shows how pointless and unreliable LLMs/AI are at even the most basic things. They've all been told that they must provide some kind of solution, always. It's especially bad now with recent updates. Claude will hallucinate 8 times out of 10 for its solutions. GPT5 now just info dumps on you hoping something in there will resonate, it can't provide an accurate precise solution anymore it just vomits on your plate and calls it a meal.

  • source
  • parent
  • hideshow 5 child comments
  • [–] 12 points 1 year ago (4 children)

    Hold up,

    [...] instead of going back to the user and saying "the font size is already set to 150%, would you like to increase it more?" it just goes ahead and bumps it up to 200 because it has to.

    No it did not. Copilot doesn't click or change anything in this video, that's the user clicking 200% after copilot told him to make it 150%.

    The only actual flaws here is Copilot not seeing the size was already set to 150%

  • source
  • parent
  • hideshow 4 child comments
  • [–] 2 points 1 year ago

    It seems like people are so eagre to bash AI that they're willing to invent their own narrative to shit on anyone who remotely sounds like they support it, even if that's not the case.

    I appreciate you actually taking the time and effort to read.

  • source
  • parent