Distill was a scientific journal, launched in 2016, that argued machine-learning research should be communicated through the web rather than the PDF. Its about page states its thesis bluntly: “Machine Learning Research Should Be Clear, Dynamic and Vivid,” and that “traditional academic publishing remains focused on the PDF, which prevents this sort of communication.” Edited by Shan Carter and Chris Olah with a steering committee of well-known researchers, Distill published a small number of unusually polished, interactive articles, many of them landmark explanations of neural-network interpretability.
In July 2021 the editorial team announced an indefinite hiatus. The post, written by Chris Olah, Nick Cammarata, Sam Greydanus, and Janelle Tam, said: “Starting today Distill will be taking a one year hiatus, which may be extended indefinitely.” It gave candid reasons. The team had concluded that publishing in a journal did less for institutional recognition than they had hoped; that the format created structural tensions between mentorship and editorial independence; and that volunteers had burned out, sometimes spending fifty or more hours improving a single article. “It is not sustainable for us to continue running the journal in its current form,” they wrote, adding that they believed self-publishing was the future for most scientific articles.
Distill’s run was short but influential. Its interactive articles set a standard for explaining hard ideas clearly, and several of its editors went on to lead interpretability research at frontier labs.
Why business readers should care: Distill is a case study in how even an admired, high-quality project can be unsustainable when it depends on volunteer labor and offers little of the career credit that the surrounding system rewards.