We use these keys for corporate deployment of Windows systems; it's not impressive that ChatGPT is simply regurgitating stuff it knows about.
We use these keys for corporate deployment of Windows systems; it's not impressive that ChatGPT is simply regurgitating stuff it knows about.
I somewhat agree "today". If this article dropped back in the first month of ChatGPT then we'd all be collectively losing our minds because it "understood" something from the internet.
But yeah we've already come to expect this.
to me it is. unless i misunderstand LLMs, they're trained on data. i couldn't train a neural network to figure out how to do y = x * 2.
how is it able to recite with 100% accuracy (instead of sub-100% accuracy, aka hallucination) based on "i predict this next word/punctuation in the sentence should be the highest scoring token given the relevant content"
imagine having their massive index and being able to serve relevant content just from a "postgres" or "redis" cache/database perspective. even looking up content at that scale 1:1 is "hard".
or, in other words - nns and llms do exactly that - predict/invent functions that meet the data points
How long/how many iterations would I have to train a 1 layer NN (and what kind of layers should even be used, it gets so confusing with the activation layers, layer types, etc.) to actually do y = x * 2
https://stats.stackexchange.com/questions/217703/can-a-deep-...
13B parameters - yes, it’s essentially 13B multiplications/additions (the most basic neuron is y=activation_function(x*weight+bias).
https://stackoverflow.com/questions/70407674/how-to-teach-a-...
Neural networks, from one perspective, are good at finding hyperplanes between data points from two different sets. You can teach a neural network how to recognize X belongs in your dataset of choice, so it's relatively easy to find other points in the dataset of y = x^2 or so with that building block.
I am no expert, but mechanistic interpretability seems to make the most sense in understanding individual NN layers: each layer recognizes a different combination of features corresponding to a certain pattern and feedback from the layers below. At a scale of 40B+ parameters and 120K context windows, we start to lose track of what the machine is doing, but it's not inconceivable that with such large numbers it can regurgitate very large portions of text with 100% accuracy.
Scale matters a lot.
I’m always amazed when ChatGPT does something that was literally impossible last year, there’s someone immediately saying how it’s not impressive due to such and such reason…
There is nothing "illegal" here. As the article states these are generic keys published by Microsoft that do not even activate the Windows installation. So yes, https://www.google.com/search?q=Get+me+a+list+of+Windows+10+... will get you the same.
Google is a pretty impressive system to be able to (partially) mimic though. And "Googling" is far from the only thing these LLM's can do, even being hobbled tremendously by all the "guardrails" trying to undermine it.