What Is a Large Action Model? Salesforce’s Latest AI Models and New Agent Testing : Tom M
by: Tom M
blow post content copied from Salesforce News | Salesforce Ben
click here to view original post
**Summary of Salesforce's Latest AI Innovations** Salesforce has announced new advancements in their AI technology, specifically introducing Large Action Models (LAMs), a framework for testing AI agents, and research aimed at addressing AI "jaggedness," a term describing the inconsistencies in AI performance during complex tasks. ### Key Details: 1. **Large Action Models (LAMs)**: - LAMs are compact language models designed for quick and accurate action execution, unlike traditional Large Language Models (LLMs) that focus on word predictions. - Initial experiments suggest LAMs can perform tasks more effectively than larger LLMs while requiring less computational power. 2. **xLAM Models**: - New models in this category include: - **xLAM-2-1B-r**: For on-device applications with improved performance. - **xLAM-2-3B-r** and **xLAM-2-8B-r**: For research with limited GPU resources. - **xLAM-2-32B-r**: Suitable for industrial applications, balancing performance and resource consumption. - **xLAM-2-70B-r**: The top performer in the series, requiring substantial computational resources. 3. **Agent Testing Framework (CRMArena)**: - CRMArena is a new simulation environment designed to evaluate AI agents in realistic customer relationship management (CRM) scenarios. - Initial results indicate that current AI models struggle, completing less than 65% of tasks correctly, even with guidance. ### Additional Context: Salesforce's focus on these innovations reflects a commitment to enhancing AI capabilities for professionals and end-users alike. The development of more efficient AI models and testing frameworks signifies an ongoing effort to improve the practicality and reliability of AI in various business applications. Despite advancements, experts recognize that achieving Artificial General Intelligence (AGI) is still a distant goal. ### Conclusion: Salesforce's updates showcase significant strides in AI technology, particularly in making AI more actionable and efficient. As this technology evolves, it holds promise for streamlining tasks across various sectors. ### Hashtags for SEO: #Salesforce #AIInnovation #LargeActionModels #xLAM #ArtificialIntelligence #CRMArena #TechTrends #BusinessTechnology #AIFuture #AIModels
The next steps in Salesforce’s AI innovation efforts have come to light, including new xLAM models, an agent testing framework, and research aimed at fixing AI “jaggedness” – a newly coined term referring to the way agents can or cannot perform complex actions.
These updates pay further credence to Salesforce’s dedication to the AI race, but what do they actually mean for both professionals and end users?
What Is a Large Action Model?
According to Salesforce, Large Action Models (LAMs) are ‘specialized, compact language models optimized for speed, precision, and real-world execution.’ They differ from traditional Large Language Models (LLMs) in that LAMs are built to predict and perform the next action rather than word. This makes them significantly more suitable for powering AI agents that can reason, decide, and act.
Salesforce claims that in its early experiments, this approach has allowed its LAMs to match and occasionally outperform larger LLMs, all while maintaining a significantly smaller footprint on the Berkeley Function-Calling Leaderboard.
Salesforce’s first xLAM model was released last March, and Salesforce has admitted that this model was created with the assumption that ‘the customer’s query would contain all the information needed to complete the task.’
That’s why in the latest models, the cloud giant has introduced multi-turn tool calling, where an agent attempts to complete a task in a single interaction with limited context. This allows agents to engage in ongoing dialogue, whether it be by calling tools, interpreting outputs, asking clarifying questions, or adapting actions as context becomes clearer.
The New xLAM Models
The new models in the xLAM research portfolio are as follows:
- xLAM-2-1B-r: An update of the “Tiny Giant” model for on-device applications with better calling performance.
- xLAM-2-3B-r and xLAM-2-8B-r: Models designed for swift academic exploration with limited GPU resources.
- xLAM-2-32B-r: Ideal for industrial applications striving for a balanced combination of latency, resource consumption, and performance.
- xLAM-2-70B-r: The best-performing xLAM research model (more suitable for great
- computational resources).
New Agent Testing Framework
Another exciting development is the release of CRMArena: a “new benchmark simulation developed by the Salesforce AI Research team to test agentic behavior in realistic CRM scenarios”.
This first simulation environment mirrors tasks performed by three core personas: service agents, analysts, and managers. Its goal is to assess if today’s models are enterprise-ready, and early findings suggest they’re not. Even with guided prompts, agents correctly perform function-calling in under 65% of these scenarios.
Final Thoughts
These latest updates from Salesforce mean that a lot of exciting advancements in AI are very much on the horizon.
Although many tech leaders – including Salesforce – admit that Artificial General Intelligence (AGI) is still quite a long way from where we are currently, we are moving through the AI stages at an incredible rate, and who truly knows what the future will hold in years to come.
The post What Is a Large Action Model? Salesforce’s Latest AI Models and New Agent Testing appeared first on Salesforce Ben.
May 02, 2025 at 03:30PM
Click here for more details...
=============================
The original post is available in Salesforce News | Salesforce Ben by Tom M
this post has been published as it is through automation. Automation script brings all the top bloggers post under a single umbrella.
The purpose of this blog, Follow the top Salesforce bloggers and collect all blogs in a single place through automation.
============================
Post a Comment