This research investigates the way that teams and communities collaborate online, with two aims: to develop new technologies to enhance collaboration and to improve our theoretical understanding of small-group dynamics. For example, one online community under study is MathOverflow, which has successfully adapted existing technology, in this case a question answering platform, to engage in problem solving in mathematics on a large scale. A detailed study of collaborations on MO revealed that mathematicians were making use of Q&A platform to broadcast problems to a large crowd, as well as commenting tools to discuss solutions and build on each others work. This work suggests that rather than shoehorning the functionality of existing large-scale collaboration platforms, systems built to support scientific problem solving should aim to strike a balance between small group discussion and large-scale crowdsourcing. In the coming years, increasing our ability to tackle scientific problems collectively, via communities like MathOverflow, will significantly impact progress in science, technology, and innovation.