It provides basically a subset (maybe around a third) of the same interesting papers that I see in conference proceedings, but it provides them earlier (typically 1 month to 1 year earlier). This can be important as my field (NLP) is quite fast-paced.
Just to give a concrete example, this paper (which was a relevant read for me) was published in TACL in July this year but was available in arXiv since February: https://arxiv.org/abs/1602.01595
Just to give a concrete example, this paper (which was a relevant read for me) was published in TACL in July this year but was available in arXiv since February: https://arxiv.org/abs/1602.01595