Reading just arXiv or even some top journal is not likely to be very helpful.
It used to be that subscribing to a journal (or ten) was the best way to keep up.
But the subscription model is dead, few academics subscribe to journals, few publisher retain print as their primary mode of distribution, and open access publications tend to get 5 or 6 times as many citations as papers that require purchase or subscription. I presume you read arXiv primarily because it is free.
I presume you are still at high school but finding that not sufficiently challenging.
I presume you have read what you school library has to offer, and need to go beyond.
I presume you have looked at what's available from your local community/university.
Been there done that... I spent my high school years doing my own physics and chemistry experiments, extracting my own reagents and building my own equipment, building electronic devices and computers, programming unbeatable AI games/programs, programs to solve problems in mathematics, joining maths clubs and winning prizes in competitions, making use of university programs designed to encourage budding scientists and mathematicians... exploring my interests in the broadest possible way!
Contact the local university and talk to people there. E.g. I'm happy to take on high school students, and indeed at Flinders University we have the Australian Science and Mathematics School (years 10-12 only) on campus to facilitate just this encouragement of gifted students - but students need to be reasonably local. Even without a formal program or relationship, academics are happy to talk to high school students, and I've run courses and competitions for high school students, given work experience opportunities and summer scholarships (not exactly restricted to local, but focussed in the state and the surrounding regions that the university draws from). Sometimes I've just responded to high school kids contacting me out the blue, by working out a program for them. My university actually has a bit of a history of this, with Terry Tao being the most famous example.
What kind of program would I recommend for someone like you... I'd start out by recommending some of the classic papers, introducing you to the originators of the seminal ideas of the field, making sure you had the basics first hand - not via the Chinese whispers of arXiv, review papers or textbooks. Review papers are often written by PhD students as part of the process of organizing their ideas - it gives them a free publication before they done any significant work on their own. It also gets them lots of citations, and some people have an h-index that is mainly about their ability to write review papers or text books. But yes, if there was a particularly good review paper or text book, I'd recommend it - unfortunately for my research area there are no good text books I can use, although I do encourage students to dip into some of those that are around - but my research oriented courses would have at least a dozen original sources of information, and perhaps half a dozen texts or reviews. This leads to an idea from Information Retrieval... Hubs and Authorities.
You've probably heard of Brin & Page's PageRank, which became the basis of Google's ranking based on how much pages linked to each other - it is closely related to singular values/vectors, looking at relatedness of pages in a symmetric way. But an earlier approach (Kleinberger's HITS/CLEVER) developed around the same time distinguished between the source and target of links, distinguishing hubs (sources of lots of links) and authorities (targets of lots of links) - these relate to the left and right eigenvectors.
What you need to find for your field is the hubs and authorities. A hub might be a review paper, or a website or society that coordinates the research area. The authorities are the people/papers that developed the seminal ideas and everyone cites. Always go to the source, cite the earliest and latest works in their developmental path. Look for the citations of that work (on Google and Google scholar, concentrating on relevant papers with hints like filetype:pdf and/or site:edu.* | site:edu | site:ac.* to get the papers by academics and find where the open access versions are lurking). A relationship with a university with relevant subscriptions would also allow you free access to IEEE, Springer etc. But all major publishers now allow people to archive preprints or postprints of their papers for free access - and this is what arXiv should be used for where allowed, but sometimes this permission is restricted to authors' own websites, or their employers'.
There are two ways to explore the field - and you can start from a single of interest. You can try to go deep and specific (citations of and by that paper). You can also go broad and applied (citations of the citations). Modern citation analysis is closely related to PageRank/HITS/CLEVER, and your library may have tools to help.