Filter Feeding

One of the most fun and occasionally overwhelming aspects of research is learning about other discoveries in the field. Conferences can help, but to avoid local bias and increase coverage, it's best to rely on a steady diet of published papers. In the first year of grad school, you might not know exactly what your field is, and the papers might look like a sea of esoteric trivia. It's like that for a while. Gradually, as you hone in on particular questions and read more deeply, the mass will start to take shape, and you'll be able to understand better how one paper relates to another.*

The cliché is that reading articles is like drinking from a firehose, and it's clearly worse if you're totally new to the field and have to catch up. (I'm also assuming you know what your "field" even is. One can pick from many firehoses.) To develop a sense of which questions are popular and how they're being answered, I think it's handy to use a RSS aggregator to manage four kinds of feeds:

1. The contents of important journals: If there's a journal that everyone in your lab reads, whose articles you always find fascinating, or whose contents are potentially immediately important (arguably, Science or Nature), you might choose to subscribe to their complete tables of contents.
2. Articles that have important keywords: A journal's impact factor can have a notoriously poor correlation with the quality and utility of individual articles published in it. It's a bad and slightly dangerous habit to think that because something was published in a so-so journal, the research is only so-so. For maximum thoroughness, set up a feed at a citation database that notifies you of any article published with a specific keyword.
3. Articles that have cited important/cool articles: If you've found an article that you think is the coolest thing ever (or wrongest thing ever, and you're working on a response), create a citation alert to notify you every time the article gets cited. It's a good idea to do this for your own articles, too, and any articles that are central to your research.
4. Articles published by an important person in your life: Basically, your adviser, though you can cite-stalk anyone. Some advisers are prolific and do not communicate very much with their labs, and you might be interested in their work. 

I think the second feed is the most important. With this approach, I stumbled on an article that mentioned an amazing and unique data set. The authors were not well known outside their country, and I'd never before heard of the journal in which the article was published. I wrote the authors and offered to collaborate, and they kindly agreed to share all their data with me. This happened in a "hot" field, and it surprised me that no one else had asked before to collaborate with them. My take-away (with n = 1) is that you can get an edge if you read broadly.

For those new to RSS feeds**, the first step is to choose an aggregator that you like. I use Google Reader, which I can open from Gmail. I use it in conjunction with Helvetireader, which is easy on the eyes and pretty perfect for skimming titles and abstracts. Popular citation databases include Web of Knowledge, PubMed, Google Scholar and Scopus. (The first and last require an institutional subscription.) If you haven't yet, do some sample searches in each to find out which is best for your field. Even if you don't have access to proprietary journals or databases, you might be able to do something useful with open-access journals, PubMed, Google Scholar, and/or arXiv.

Currently, my feed gets roughly 200 articles/week. I'm not sure this is good, but one of my research subjects is very popular. For the vast majority of the articles, I don't read past the title; with a subset, I read the abstract very closely; and for the remaining few, I read the abstract and article itself (to some degree). Figuring out the right ratio of skimming to reading has been a big challenge for me. It would be interesting to see how scientists' reading habits vary.

*To a certain extent, these shapes are always changing--it's hard to know what methods or analogous systems might inform the questions we're pursuing.
**I'm not sure if it would be helpful for me to write a post that outlines this process in detail. Please let me know if you'd like one.