Yandex has trained a neural network to check its own texts for hallucinations. This approach allowed it to reduce the number of inaccuracies in short descriptions of places and attractions in Maps by six times, the company’s press service told RB.RU.
Author:
https://rb.ru/author/ntihonov/
Subscribe to RB.RU on Telegram
YandexGPT creates short descriptions in several stages. First, it analyzes reviews about the place and the organization card filled out by the owner on Maps, and then combines what is important to visitors and the characteristics of the place into a short story. By removing too general formulations and “water” from the text, it creates a description with a given style and structure. But when preparing descriptions, the neural network sometimes adds non-existent details – they are called hallucinations.
Previously, five different models were responsible for creating such descriptions, but now only one does so and it requires less computing power to run. These changes increased the veracity, precision and quality of the summaries by 90%. It was also possible to improve the quality of the neural network thanks to additional training.
“Yandex expanded the dataset with reference examples from previous generations of models – this is how the YandexGPT-based model learned to create more useful descriptions. For example, now, when preparing texts, it less often uses common phrases like “delicious food” or “pleasant atmosphere,” the company says.
They added that several hundred thousand cards with a brief description are already available on Maps.
Yandex Maps also updated the section with personal recommendations for walks and entertainment in mid-August. Ideas can now recommend parks, cinemas, museums, concert halls and other places.
Author:
Nikolai Tikhonov
Source: RB
I am a professional journalist and content creator with extensive experience writing for news websites. I currently work as an author at Gadget Onus, where I specialize in covering hot news topics. My written pieces have been published on some of the biggest media outlets around the world, including The Guardian and BBC News.
