Oh yes. This is going to be glorious to watch. Before, prompt injections were mostly a joke. With this, they will truly become a proper, new type of attack vector.
This sounds like a massive security hole to me, especially if this runs in the cloud.
Is this thing sending _all_ context of the programs on your windows computer to MS?
I'm also a bit curious how they will actually pay to have millions of these LLMs running and how much co2 it will output. This doesn't exactly fit their narrative about the new power saving features in their consoles.
Having spent most of my ChatGPT time trying to get it to correctly write powershell/batch scripts to do different things in Windows, I will be fucking amazed if this button actually does half the things you tell it to.
In the video the user drag&drops a file into the copilot window. How do you know that it can see your screen? I believe "works with all apps" only refers to the third party plugins.
You: Clippy, Open the CMD window please.
Clippy: I'm sorry, I'm afraid I can't do that. This integration is too important for me to allow you to jeopardize it.
A new permanent taskbar button opens an OpenAI LLM trained to use Windows 11.
You ask the LLM to do what you want -- "summarize this email and send the summary to my boss."
It works with all applications. It sees what you're seeing.
As a longtime Linux user, it pains me to say this:
It's... beautiful.
...
But it's also scary, because it's not under your control. Ultimately, it obeys only the bureaucracy at Microsoft.
In that sense, it's like "MOTHER" in Ridely Scott's Alien and "HAL" in Kubrick's 2001: A Space Odyssey.