I'd refine that second statement: it's not usually in a graduate student's best interest to collaborate on a project, but the student's PI typically has a competing interest (and a longer time horizon), and so collaborations still happen. The smart PI just manages the risk by assigning collaborative projects to his students -- he gets to reap the rewards of the rare good collaborations, whereas the bad collaborations merely crush the unfortunate grad student to whom they are assigned...and nothing of value is lost.
Part of the game of grad school is recognizing and fending off the "collaborators" who will do nothing but slow you down and/or take credit for your work, while still remaining open to that rare collaboration that can make your career -- the two most successful graduate students I know basically walked into career-defining projects as a result of collaborations that were handed to them in their first year.
Of course, you can be torpedoed by a bad collaboration at any stage of your career, so this isn't a dilemma unique to graduate students. Collaborations are useful only so long as you're getting more than you're putting in.
Part of the game of grad school is recognizing and fending off the "collaborators" who will do nothing but slow you down and/or take credit for your work, while still remaining open to that rare collaboration that can make your career -- the two most successful graduate students I know basically walked into career-defining projects as a result of collaborations that were handed to them in their first year.
Of course, you can be torpedoed by a bad collaboration at any stage of your career, so this isn't a dilemma unique to graduate students. Collaborations are useful only so long as you're getting more than you're putting in.