Founded in April 2024 by Marvin Purtorab and Andy Toulis, two machine learning engineers who met at Shopify, the e-commerce platform where they worked on recommendation systems and AI assistants, Convergence brings together a team of Google alumni DeepMind, Meta, OpenAI and PolyAI.
Mundane tasks
By pairing users with personalized AI, Proxy can learn tasks and workflows; which frees workers from their administrative burden. To put things into perspective, 62% of the average employee’s workday is wasted on repetitive, mundane tasks, which Proxy is designed to excel at. Over time, the agent will take care of mundane tasks, allowing you to focus on more important tasks. “If you look at the current landscape, you will see that a large number of companies are building these types of agents as vertical agents: a sales agent, a human resources agent…” notes Marvin Purtorab. “Our view is to take a different approach. We’re trying to lay the groundwork for the first general class of agents that can, depending on the user, multiply into whatever type of agent you need and do the things for you that you don’t want to do. personally.“
“There is an urgent need to act!”: The MIC and Microsoft want to train 300,000 Walloons in generative AI within three years
Unlike most agents designed for specific workflows, Proxy’s differentiator therefore lies in its ability to work on a range of tasks and domains, learning skills like a human would through memory. long-term and continuous learning. This is made possible by a new class of models called Large Meta Learning Models (LMLM), who are trained to gain the ability to learn on their own, a growing field of artificial intelligence that aims to create models that can quickly adapt to new tasks with minimal data . They are trained specifically on a diverse set of tasks to learn how to manage (i.e., remember, store, and delete) items in and from their memory; which makes it possible to extract common learning strategies and patterns.
