Salesforce 对谈哈佛商学院教授：大企业如何做创新

The striking number is the ablation: removing teacher review and code validation halves accuracy, suggesting the guardrail pipeline matters more than the skill library itself. Teams building production LLM agents can borrow the pattern of logging every corrective intervention as a validated, reusable program. Compared with plain ReAct, HASP trades a one-time curation cost for roughly 15% token savings and a 20-point accuracy gain on web-search reasoning.

官方账号Salesforce AI Research@SFResearchSource

Building agents that model actual human behaviour has relied on synthetic personas, and this benchmark offers a measurable, open alternative with licensed data. Teams developing personalised assistants or recommendation agents can use it to evaluate how well their models exploit user history. The Apache-2.0 release on Hugging Face makes it directly usable without licensing friction.

官方账号Salesforce AI Research@SFResearchSource

Verified fitness gates, not raw model skill, are what turn LLM-generated code edits into production-safe optimisations — the correctness pipeline is the real contribution here. Engineering teams maintaining legacy codebases, especially Salesforce Apex and Java shops, can apply the runtime-validated search pattern to automate hot-path optimisation without shipping regressions.

官方账号Salesforce AI Research@SFResearchSource