In 2010, MIT Media Lab students Karthik Dinakar and Birago Jones embarked on a project to develop a tool to aid content moderation for companies like Twitter and YouTube. The original concept had difficulty identifying posts using teenage slang and other non-obvious language nuances, but after realizing those most familiar with the data should be the ones building machine-learning models, the pair developed point-and-click tools that empower nonexperts to do just that. This led to the formation of Pienso, a start-up helping users build large language models capable of detecting misinformation, human trafficking, weapon sales, and more.
Jones and Dinakar started their work on Pienso in 2010 while studying as graduate students. Despite its early success, including invitations to the White House, they delayed full-time work on Pienso until 2016, when Dinakar completed his PhD and deep learning began to garner more interest. Today, Pienso has several prominent partners, including SkyUK, whose customer success team uses the tool to model and understand customer complaints effectively, processing half a million calls daily.
In 2020, government officials sought Pienso’s help in understanding the emerging Covid-19 crisis. The tool enabled experts in virology and infectious disease to create machine-learning models to analyze thousands of research articles on coronaviruses, which later helped identify and support critical supply chains for drugs.
Pienso’s flexibility, running on both internal servers and cloud infrastructure, offers an alternative to businesses that would otherwise have to share their valuable data with other AI companies. Its partnership with GraphCore, announced earlier this year, is expected to increase efficiency and reduce latency even further. The Pienso platform is likened to “an Adobe Photoshop for large language models,” enabling data import, refining, annotation, and analysis in an intuitive and user-friendly manner.
Jones and Dinakar believe that Pienso is supporting the future of AI by allowing individuals most familiar with the data to develop models for their specific needs. They envisage a future where a ‘garden of collaborative models’ are orchestrated by the people who best understand the data, offering a more tailored and effective solution than a single all-encompassing model.
