“`html
Synthedia’s Regular Readers Will Be Familiar With the Difference Between Knowing and Doing Assistants
The latest in generative AI-enabled assistants knows many things, fulfilling knowledge-based tasks at an unprecedented level. Voice assistants like Alexa, Google Assistant, or Siri excel at controlling applications.
MultiOn Agents
The Information recently reported that MultiOn will announce a $20 million funding round at a $100 million valuation. This rapid rise just one year after founding and less than six months after its first product launch is noteworthy.
General Catalyst, Forerunner Ventures, and Blitzscaling Ventures are leading the round, with Amazon’s participation in both rounds indicating their interest. A person involved in the funding round mentioned discussions about MultiOn’s agent tech potentially enhancing Alexa.
From Knowing to Doing
Large Language Models (LLMs) excel at determining user intent but struggle with executing tasks reliably within applications. Agents, like those developed by MultiOn and other platforms, bridge this gap by combining LLMs for understanding with specialized agents for execution.
This approach allows the LLM to interpret a task request and then delegate it to the most appropriate agent, ensuring both knowledge (what should be done) and action (how to do it) are handled.
MultiOn Examples
MultiOn launched a playground in late May 2024 with several examples of agent capabilities, including adding items to an Amazon cart, looking up weather, reserving tables at restaurants, and summarizing news. These tasks are not novel but represent the application of AI agents in fulfilling them.
Rabbit Runs, Agents Negotiate
Recently, Rabbit announced its large action model (LAM) to perform similar tasks. However, recent investigations suggest Rabbit is not using an LAM but instead employs Playwright automation scripts for app control.
The team’s decision to use hardcoded procedures rather than a more flexible AI approach may indicate challenges in implementing the intended LAM technology or a strategic choice based on current capabilities.
Key Takeaways
- Agents like MultiOn and Rabbit are critical for bridging the gap between knowledge-based tasks and application control.
- The use of agents allows for more flexible task execution, leveraging both LLMs for intent understanding and specialized agents for action.
- Developers should consider the practicality of agent approaches versus imperative programming models like Playwright scripts to ensure robust and adaptable solutions.
“`
Originally published at synthedia.substack.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.