This article was published on August 18, 2016

Facebook open-sources its AI-building research so you’ll stop building shitty bots


Facebook open-sources its AI-building research so you’ll stop building shitty bots

Facebook wants to show you how the sausage is made.

Facebook is open-sourcing its AI bot-building research. Part of its Facebook AI Research (FAIR) lab’s mission, the plan is to provide better access to researchers and engineers to they, in turn, can build better bots.

The code library, called ‘fastText’ is now available on Github and according to Facebook, only requires a compiler with “good C++11 support.”

The appeal of fastText is in the speed  and efficiency at which you can train your bot. FAIR claims you can train models “on more than 1 billion words in less than 10 minutes using a standalone multicore CPU.” Compared to deep learning models, fastText could cut training time from days to seconds.

In a statement, FAIR said:

Ultimately, we hope that fastText will help us all design better applications and further advance the research in language understanding.

After integrating chat bots in its Messenger app earlier this year, it’s easy to see why this move makes sense for Facebook. As we’ve detailed extensively, most bots just flat-out suck. If Facebook hopes to obtain any level of success with these integrations, it might be time to pony up the tools to help developers create them.

And that looks to be just what it’s doing.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with