iMerit opens a platform with 30,000 experts to test AI models
The company says that human experts who try to break a model find mistakes that automatic tests can miss.
iMerit, a firm that labels data and gives AI training services, started a platform with the name iMerit Scholars. It helps AI developers hire and manage experts who test, tune and check models for hard business uses. The group has more than 30,000 checked experts in 118 countries. ExlService bought iMerit in July.
How it works
The platform gives one set of tools to recruit, train, test, give work to, and pay experts. It uses the Ango Hub tool of iMerit for data labeling and model tuning. Trolley Technologies does payment and compliance. iMerit manages each project and does quality checks and audits.
The numbers
- The group has 106 languages and fields, such as medicine and farming.
- A test has three levels: general AI skill, field skill, and inspection by other experts.
- Prices are for each hour, for each deliverable, or for each measured result.
- One customer asked for about 400 more workers in three or four days.
What is not known
The company gave no test results that show better models. It gave no prices. The plan for hundreds of thousands of experts is not a fact.
Insight: Two ways to check a model are growing apart
IBM released a plugin that lets developers monitor and change a model while it writes. It adds 2% to 6% more time. iMerit sells experts who “torment” models because automatic tests can miss errors. This reads as two types of model check: software in the engine, and human experts external to it. Regulated teams will maybe pay for the two.
Sources
Posted
