1 min read
Add as a preferred source on Google

Meta’s Agents Complete 1,097 Experiments, Lift Search Sessions 0.23%

The 31-day campaign used an estimated 360,000 GPU-hours. The combined model also slowed search-interface rendering, but the size of the slowdown was not disclosed.

An anonymous hand adjusts a search-results card beside experiment cards / TokenPost.ai
An anonymous hand adjusts a search-results card beside experiment cards / TokenPost.ai

Meta’s AI agents completed 1,097 candidate experiments to improve its search-suggestion system over 31 days, and a live test found the combined model lifted search-session volume by 0.23% compared with a control model.

The estimated campaign consumed about 360,000 GPU-hours, including additional tests and failed or retried jobs. The resulting model also slowed search-interface rendering, though the size of the slowdown was not disclosed.

The agents proposed model changes, wrote code, submitted training jobs and screened results. Meta engineers set the research goals, steered the search and reviewed the combined model before it was tested on live traffic.

Of the 1,097 completed candidate experiments, 325 advanced based on offline results. The live test evaluated one combined model and recorded a 0.23% relative increase in search-session volume, with a 95% confidence interval of plus or minus 0.15 percentage points. Other engagement measures, including participation and retention, also moved in a positive direction.

The research paper does not resolve whether Meta will permanently roll out the model or how its results compare with a matched effort led only by human engineers.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…