Salesforce's AI agents improved one of its AI models in about 80 hours, it said. Its scientists often need several weeks.
Salesforce described the work in a post on 11 November.
The agents are given one goal, a system to work on and a computing budget. They form hypotheses, run experiments and keep a change only when it shows a verified improvement over the baseline, the company said.
The model they improved is the one Salesforce uses to find the right records in Data Cloud, its data product. On 10 public test tasks its score for ranking relevant results rose by 6.13 points. A second session of roughly nine hours brought 36% more throughput and 17% lower latency, meaning more requests handled and shorter waits, with almost no loss in quality, Salesforce said.
"We used to count progress in weeks of human work. Now we count it in hours of compute," a Salesforce research director said.
Customers do not have the new model yet. It still has to pass the checks required for live use before any customer sees it, the post says. Researchers set the goal, the budget and the data limits.
The desk's reading: the post gives no cost for the 80 hours of computing and no date for customers. It also does not say how many researchers the weeks-long cycle used to occupy.
Sources
Our file on Salesforce
- 13 Nov
ServiceNow released AI it says moves firms off rival sales software in weeks. It named no price or customer that switched.
- 11 Nov
Salesforce reported its AI sales agents save sellers an average of six hours a week. Its survey covered 24 customers.
- 5 Nov
Salesforce said Coworker had more than 185,000 users. CIOs still need savings evidence Salesforce has not supplied.
- 4 Nov
Half of Slack messages are no longer from people, Salesforce says. It counts a billion messages a day.
- 3 Nov
Salesforce agreed to buy Cresta, whose AI agents answer customer calls for Marriott and Comcast. It gave no price.
For your next steering committee
- 1Ask which AI tools already touch your ERP or finance data, including the ones nobody approved.
- 2Ask what each tool is allowed to change, and who reviews what it changed.
- 3Ask how the result is measured, and who decides to switch a tool off.
Issue 01 of the ERP LEADERS brief puts a timeline, the three-number review and six questions on one page. Read issue 01
Welcome back. · Issue 01
