KI model at test level: one setting applies to all

Erst vor wenigen Monaten hat Meta sein neues KI-Modell Llama 3.1 veröffentlicht. Artificial intelligence and Open-Source are within the reach of researchers: they must solve problems during difficult experiments. So this is a group of Forscher who leads a group from Colorado Reed, and who is also an applied research scientist at Apple.

Ziel des Experiments war est, herauszufinden, wie leicht sich KI-Modelle aus der Bahn werfen lassen können. The New Scientist is ready to model its own parameters with billions of precise parameters. Often this is the Forscher: inside an internal space of the KI models to use.

There is a parameter in a KI model to know what it is

To find them, the researcher chose the generated Llama-7B model. As the name suggests, the KI has a Milliarden parameter. This information is provided by the generated text, which will be provided by a prompt. Im Grunde stellen die einzelnen Parameter Wahrscheinlichkeiten für bestimmte Textbausteine ​​​​dar, die die KI then aneinanderreihen cann.

Die Forscher: innen haben cese Zahlenwerte verändert, um herauszufinden, wie sich der Output verändert. However, if the old Parameter has fallen, without a single upgrade, the Output will be damaged. Dem entgegen stehen aber sogenannte Superparameter, the besonders wichtig sind. One of these parameters is used, generated by the KI model in our hallucinations or in the Kauderwelsch totality.

Im Test veränderten die Forscher: innen rund 7,000 Parameter without Auswirkungen auf den KI-Output. When the superparameter is selected, the Llama-7B does not have more features. He created Yingzhen Li from Imperial College London and this setting was highlighted by KI models. If you do not have any parameters, you will have to set the following parameters to the next parameter.

And Thomas Wolf, the director of Hugging Face, is the best manager of the teams. As of 2017, this is a large model of OpenAI’s model, with 82 million people trained by Amazon. Die KI sollte nach dem Training imstande sein, negative et positive Bewertungen zu unterscheiden. Laut Wolf proposes this definition of the parameters of a parameter in the KI ab model.

Please note that this Erkenntnis für die Weiterentwicklung von künstlichen Intelligence rentieren. Denn KI-Forscher: innen müssen Wege finden, wie Modelle auch ohne cese besonders wichtigen Parameter auskommen etwa andere Wege zu ainer richtigen Answer finden können.

So I invite you to a local KI