In 2010, Media Lab students Karthik Dinakar SM ’12, PhD ’17, and Birago Jones SM ’12 wanted to develop a tool to assist content moderation teams at companies like Twitter and YouTube. The project prompted excitement, earning them a demo at a White House cyberbullying summit. When Dinakar struggled to create a working demo, Jones realized their model could not interpret teenage slang and other indirect languages. This made the team understand that the builders of these models should be the individuals who best understand their data and led them to develop the Pienso platform.
Pienso offers point-and-click tools, allowing non-experts to create machine-learning models. These are being used to construct large language models to identify misinformation, human trafficking, and weapons sales, among other things. The two scholars developed Pienso while graduate students in the MIT Media Lab’s Software Agents research group.
Enthusiasm for the project resulted in further invitations to the White House. Once Dinakar finished his PhD at MIT in 2016 and deep learning grew in popularity, the duo dedicated themselves full-time to Pienso. The scholars attribute their success to the exposure and training they received at MIT and the institution’s Industrial Liaison Program (ILP) and Startup Accelerator (STEX), which connected them with early partners.
SkyUK, one of the early partners, used Pienso to understand their customer’s most common queries. The models constructed through the platform were later used to manage half a million customer calls daily, saving the company £7 million pounds by shortening call length.
When the Covid-19 crisis hit in 2020, Pienso’s founders were contacted by government officials. The tool was used to set up machine-learning models to examine thousands of coronaviruses research articles, record findings, and fortify critical supply chains for drugs.
Pienso, compatible with internal servers and cloud infrastructure, is like an “Adobe Photoshop for large language models,” according to Dinakar. It simplifies refining data, preparing it for deep learning, analyzing and structuring. The tool can create fine-tuned, large language models in 25 minutes without writing a code line.
Recently, Pienso partnered with GraphCore to provide a more effective computing platform for machine learning. This partnership will reduce latency and make AI more accessible. Pienso’s founders believe their solution will enable more effective AI models tailored to address specific problems by those who understand the problems better. They plan to integrate multiple AI models to work together with the best understanding of the data.