I'm glad to see I've made you curious! On this page you'll find interactive demos of projects that you can run in your webbrowser to your heart's content. If you have any questions, feel free to message me.
This card showcases an answer localization model: ask any question about a portion of text, and if the answer is present in the text, you will see exactly which words in the text answer your question.
Unlike generative models like ChatGPT, this model does not hallucinate facts. Instead of generating text, it locates an answer within the text that is provided, making the output consistent and predictable, and thus more useful for business applications. Furthermore, at 100 MB it is orders of magnitude smaller than the aforementioned models, which makes it possible to run client-side on any smartphone these days. This model works by using a compressed version of BERT for Natural Language Processing (NLP).
BERT
To get answers from text, first (down)load the model of 100 MB (careful if you're on mobile data). Type the question you'd like to answer (or keep the suggestion) and press the "locate answer" button. You should now see where in the customizable text your ansewr lies. Feel free to experiment as often as you like with your own questions, or even your own passage of text!
100 MB