DeepL Unveils New Language Model Offering Superior Translations Compared to Google Translate and ChatGPT
DeepL has introduced a new language model to power its translation system. The model is reported to outperform both Google Translate and ChatGPT.
DeepL Unveils Superior Translation Model
Since 2017, DeepL has been one of the toughest competitors to Google Translate in machine translation. Now, the Germany-based startup has introduced a brand-new large language model (LLM). The new model is said to outperform many of today’s translation systems.
The new language model is described as being specially developed to focus on translation and editing.
It is claimed to perform better than Google Translate and ChatGPT.
According to the company’s statements, the model’s aim is to produce more human-like translations and reduce the risk of hallucinations. Hallucinations are a significant problem in AI models, referring to when models produce outputs that are false or incorrect.
Unlike its competitors, the model was not trained using internet data. Instead, DeepL uses its own data. Additionally, language experts have trained the model to enhance translation quality.
The company claims that the model outperforms leading translation companies like Google, OpenAI, and Microsoft. This means it is better than the widely used Google Translate and the recently improved ChatGPT. Tests showed that Google Translate requires twice as many edits to reach the quality of DeepL’s model, while ChatGPT needs three times as many edits.
In addition, tests were conducted among their own models. The results showed a 1.7-fold improvement in English, Japanese, and Chinese, and a 1.4-fold improvement in English-German pairs.
The system with the new model is currently available only to DeepL Pro subscribers for four languages: English, German, Japanese, and Chinese. Users can select and use the next-generation model via the web.
.
Page Contents
Toggle