Artificial Intelligence - Start-up DeepL introduces new AI model for translations
German AI startup DeepL has introduced a new generation of its language model (LLM) to the market to improve the quality of its machine translations. The company based in Cologne referred to blind tests with language experts, which showed that DeepL translations based on the new "Large Language Models" were preferred more than the results from systems of large US internet corporations.
LLM are machine learning models for Artificial Intelligence trained to understand and generate human language. Well-known LLMs include GPT-n from OpenAI, Gemini from Google, or Llama from the Meta Platforms Inc. The new model from DeepL is based on an LLM technology specifically developed for translation and text processing, explained the Cologne startup.
Unlike general models that train on freely available internet data, DeepL uses specific data for training its LLMs, the company emphasized. These data have been fine-tuned for content creation and translation for seven years. To ensure the model delivers the best possible translation, DeepL also employs "thousands of manually selected and specially trained language experts" to optimize it.
According to the company, blind tests have shown that DeepL translations were preferred 1.3 times more often by language experts than Google Translate, 1.7 times more often than ChatGPT-4, and 2.3 times more often than Microsoft. Translations based on the new model are available to DeepL Pro customers for English, Japanese, German, and simplified Chinese from the start. More languages will follow soon.
DeepL Pro is the premium translation solution for businesses and organizations that need to translate frequently and want to protect their data. The company also offers a free DeepL variant for short text translations.
After a financing round in May, DeepL is valued at $2 billion (€1.85 billion). The most valuable AI startup in Germany recently expanded its product portfolio with a writing assistant with Artificial Intelligence (DeepL Write Pro). DeepL employs around 900 employees and staff.
- DeepL's new LLM for improvement in machine translations has gained preference over systems from large US internet corporations like Google and Facebook in blind tests conducted with language experts.
- In the realm of AI, LLM are machine learning models that have been trained to comprehend and produce human language, with notable examples being GPT-n from OpenAI, Gemini from Google, and Llama from Meta Platforms Inc.
- For its LLM training, DeepL focuses on specific data that have been finely tuned for content creation and translation for a period of seven years, setting it apart from general models that rely on freely available internet data.
- To ensure the highest quality translations, DeepL employs a vast number of manually selected and specially trained language experts to optimize its LLM.
- DeepL Pro, the premium translation solution for businesses and organizations, offers translations based on the new LLM technology from the start for English, Japanese, German, and simplified Chinese, with more languages expected to be added soon.
- Recently, DeepL, the most valuable AI startup in Germany, expanded its product portfolio with a writing assistant powered by AI (DeepL Write Pro), demonstrating its commitment to the field of Information Technology and Software.
- With a workforce of around 900 employees and staff, DeepL continues to innovate in the areas of Computer Science, Artificial Intelligence, and Linguistics, contributing significantly to the IT sector in Germany, particularly in the city of Cologne.
