Emergence’s Post

Welcome to Emergence, where the future of enterprise workflow automation begins. Listen to exciting words from our co-founders Satya Nitta, Sharad Sundararajan, and Ravi Kokku, Learn Capital's founder and investor Rob Hutter, our research scientist Ashish Jagmohan, and our Chief Design Officer Hélène Alonso as they share how we’re advancing the science and development of #AIagents. Follow us to discover how intelligent agents will unlock the full potential of #AI in enterprise systems.

Our Principal Engineer, Tamer Abuelsaad, shares how AI agents self-improve by "skill harvesting", a concept we created. Follow us to learn about how we're advancing the development of self-improving systems for enterprises. #SelfImprovingAI #AIagents

Three must-reads on agentic systems, written by Emergence’s engineers: 1️⃣ Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems 👉 https://lnkd.in/d8jyXdxG 2️⃣ Multimodal Auto Validation For Self-Refinement in Web Agents 👉 https://lnkd.in/ez2Tupny 3️⃣ SEAL: Suite for Evaluating API-use of LLMs 👉 https://lnkd.in/e8yn8jYY #AIAgents

Thrilled to be among the 10 enterprise AI startups featured at #LlamaLounge 14! Stay tuned for more details 👀

Skill harvesting is key to allowing agentic systems to perform efficiently, and here's why👇 Agent-driven systems are created with a set of predefined skills that can carry out specific tasks within digital environments. At Emergence, we introduced a primitive-skill-based web navigation functionality for our state-of-the-art web navigation agent, Agent-E. Agent-E uses "skill harvesting", which allows it to self-reflect, thereby autonomously developing more specialized higher-order skills from its primitive skills. This enables the agent to perform its tasks in a highly efficient manner. Take a look at the improved efficiency in time and steps taken before and after skill harvesting in Agent-E. Read more about skill harvesting here 👉 https://lnkd.in/eTZkhKxj #AIAgents #AgentE #SkillHarvesting

Planning and grounded tool use alone aren’t enough for enterprises. Here are the key characteristics needed by AI agents to guarantee reliability and navigate stochasticity: 1. Hierarchical and dynamic planning: This enables separation of abstractions at different levels of planning, where the highest level is more human comprehendible, and the lowest level should be tool comprehensible. In the event of the plan’s failure (whether due to the creation of a wrong plan or changes in the environment), the planner can replan and attempt to complete the task. 2. Verification and Reflection: To operate in a highly stochastic and dynamic environment, an agent should be able to constantly monitor the output quality of every step of the plan. We are building high-quality multimodal verifiers that can use diverse signals to ensure robust verification. 3. Self-Improvement: An agent can never be 100% performant from the start of its operation in an enterprise workflow. The ability of an agent to adapt to the environment and self-improve with time is crucial for them to be useful for enterprise tasks. We are using techniques such as auto-curriculum learning, skill harvesting, iterative refinement, critiquing, and multi-step reasoning to create self-improving agents. Curious to learn more? Keep an eye out for our upcoming research papers on AI agents, including multimodal verification, self-play in web agents, and #API and tool calling test bed. #AIAgents #SelfImprovingAgents

Interested in how to encode specific innate knowledge in modality-agnostic AI systems? Read our latest paper: "The Ungrounded Alignment Problem." Co-authored by our researchers, Openmind Research Institute and Sakana AI, this paper shows how selected innate knowledge could be embedded into highly plastic AI systems. It highlights the potential to create AI with specific drives or goals while maintaining flexibility. Dive into it now 👉 https://lnkd.in/efGJ2CBH #AI #AIagents

“Self-improvement feels like an important milestone, and a truly transformative milestone, to making AI systems that are near human intelligence.” Hear from our Research Scientist, Ashish Jagmohan, as he explores how we’re building self-improving AI agents that closely mirror human cognitive abilities. Follow us to learn more! #AI #AIAgents #SelfImprovingAgents

Thrilled to announce that our CEO and co-founder Satya Nitta will attend United States Energy Association 's virtual press briefing series on 2nd October 2024, ‘The AI Revolution Underway in the Utility Space.’ Register today👇

Very interesting observations on diffusion transformer layers by Naga Sai Abhinay Devarinti, following the approach presented in our recent paper. Read more about our research on the internal workings and functional roles of transformers, developed in partnership with Sakana AI 👉 https://lnkd.in/e2xUpz2m