The renowned product AI and research lab Adept recently announced that they came from nowhere and now have raised about $65 million in the Series A round which was led by Greylock and Addition. This round included many other participants from Angels and Root Ventures like Howie Liu (Airtable founder), Andrej Karpathy (Tesla Autopilot head), Chris Re (Stanford), Sarah Meyohas and Scott Belsky (Behance founder).
Adept was first founded by a group that had expertise in large-scale neural networks. They were so good, that David Luan, the CEO of Adept even led the large model program by Google, before which he was OpenAI’s head of engineering. The CTO and Chief Scientist, Niki Parmar and Ashish Vaswani, respectively, also worked their fair share at Google. They were Google Brain researchers and invented the Transformer, a revolutionary architecture that changed the way computers understand natural language forever. Since this happened in 2017, Transformers have been able to resolve decade-old problems helping power significant applications like GPT-3. This model can generate human-like languages like emails, trivia questions and tweets.
With the huge success of Transformers, all three then-Google employees, Niki, David and Ashish started training bigger and larger Transformers. They had one single objective in mind which was to gradually develop a model that could help power all ML use cases. However, eventually, with this process, they discovered a considerable limitation. They saw that models such as GPT-3 could write amazing prose but lacked the ability to take action in the digital world.
As David Luan put it, all Transformers and their applications embody the first and largest step toward general intelligence. However, their group believed that real general intelligence needs Transformers which can also act, instead of just reading and writing. At Adept, they are building a model that uses all API and software tools that most people are using today.
Users are working alongside Adept’s new technology, utilizing this natural language interface on already existing software like Photoshop, Airtable, Twillo, Tableau and ATS. The company thinks this is a more convenient approach to developing general intelligence that makes users the main drivers, enabling them to free up time for tasks they enjoy most by offloading all manual tasks.
There is no doubt that machine learning has come a long way since its inception and Adept’s team has a lot of influence on it. They have basically invented it, while also having a significant role to play in several other breakthroughs. While there are several a few more start-ups that have the same technical ability as Adept, no one has invested and come out successful as they have. Greylock is now excited to see what Adept’s next move is and how it will advance this industry even further. As Greylock partner and Adept board member, Saam Motamedi said, Adept is capable enough to push the boundaries and build more efficient models while also increasing business and user value