Salesforce takes aim at ‘jagged intelligence’ in push for more reliable AI
1 min read
Summary
Salesforce has announced several updates to its AI offerings, with a focus on bridging the gap between artificial intelligence and unpredictable enterprise environments, referred to internally as “jagged intelligence”
Among the updates is a new benchmarking framework called CRMArena, which simulates real-life customer relationship management(CRM) scenarios, enabling the comprehensive testing of AI agents in professional contexts.
Also unveiled were new embedding models, such as SFR-Embedding, which has been trained on massive amounts of data to better understand context; and “xLAMs”, a family of models designed to predict actions, rather than just generating text, and which are tuned to out-perform on specific, enterprise-level tasks.
Salesforce is also enhancing trust and safety within its AI offerings, with a family of models called SFR-Guard, which is trained on both publicly available and CRM-specialised internal data to provide guardrails for AI agent behaviour; and a benchmark for evaluating judge models in context called ContextualJudgeBench.
The company highlighted its focus on consistency and reliability over raw intelligence as it takes on Microsoft, Google and Amazon in the race to dominate business AI.