Users of the crowdsourcing platform "Yandex Tasks"complainedfor low pay for retyping text from documents - it was 10 kopecks. The author of the tasks is a document digitization startup OOO "DВrain" (DВrain, a Skolkovo resident). As noted by users of the platform, they have encountered similar tasks for an even lower cost - one, two or three kopecks.
The press service of DВrain told a ComNews correspondent that DВrain places tasks for document digitization on third-party crowdsourcing platforms in addition to recognizing cases that artificial intelligence could not handle: "For example, on printed registration stamps in a passport, our AI recognizes 95% of stamps. For one reason or another, we cannot make out the remaining 5% (very hard to see, etc.). If the customer needs 100% recognition and agrees to involve taggers (people who help determine what is written in hard-to-read fields), we go to Yandex Tasks. Or to Toloka, as it was before. Some clients verify unrecognized data within the company, involving an employee from within. Some want to give us the full cycle."
In 2023, the crowdsourcing platform Toloka separated from Yandex, went to the international market and is now called Toloka AI. But the Yandex Tasks platform appeared - it began to work on the Russian market.
"Yandex Tasks" works on the following principle: a company places tasks on it, for example, DВrain places a cut-out piece of a document, and a person who wants to complete the task and earn money must enter the text that he saw in the picture into the field.
When asked by a ComNews correspondent what stage of development the DBrain AI model is at, whether it exists at all or all documents are deciphered by people for portugal whatsapp resourcea fee, the DBrain press service replied that the startup guarantees its clients that the vast majority of documents are recognized in up to 5 seconds - during this time, tasks will not even have time to be created in Yandex Tasks, respectively, DBrain AI models exist: "There are dozens of such models: they determine the type of document, search for fields, translate images into machine-readable text, search for fakes in documents. And only if these algorithms have doubts, we rarely send part of the data for manual verification. DBrain has been developing solutions for extracting data from documents since 2019, we have an ML department that trains and develops our models for working with documents. This is an in-house solution."
"Yandex Tasks" is a two-way marketplace, like, for example, YouDo or "Profi.ru". The author of the task sets the price, and the performers take or do not take the tasks.
"DBrain has an internal algorithm that controls the price depending on the number of tasks, the required speed and quality. Previously, a task on Toloka AI could not cost less than one cent. We considered this price to be very high for some tasks, but it was technically impossible to lower it. The task itself is to click "yes" or "no" or retype one or more words from a picture. Then we wrote an integration for moving to "Yandex Tasks". When we moved there, our price automatically moved - only instead of one cent, it became one kopeck," explains the press service of DBrain.
"We gave the algorithm time to figure out the price, and we ourselves observed what would come of it. As expected, people began to write that it was a small amount of money. But they still continued to take tasks, although they could have not done so. Then the price increased 10 times. The minimum price for our tasks is now five kopecks, and for the simplest question, where you need to click "yes-no". And we pay 30 kopecks for a set of questions. If the quality and speed still do not satisfy us, the price will increase itself," added the representative of the DBrain press service.
How data is anonymized
How is the rate for completing a task formed?
-
- Posts: 679
- Joined: Thu Jan 02, 2025 7:05 am