DeepL launches new AI model for translation and text editing that supposedly beats GPT-4
Key Points
- DeepL has unveiled a new large-scale language model (LLM) for text translation and editing that is more preferred than solutions from Google, OpenAI and Microsoft, according to internal blind tests.
- The specialized LLM combines DeepL's proprietary training data with feedback from thousands of human language experts to deliver more human-like translations and reduce the risk of hallucinations and misinformation.
- The new model is initially available in selected languages for DeepL Pro customers. Further details on functionality, testing methodology and training data are not being disclosed by DeepL.
German AI company DeepL has unveiled a new specialized large language model (LLM) for translating and editing text. The company claims this update sets a new industry standard for translation quality.
According to DeepL, the new model combines three key features:
- A specialized LLM optimized for language tasks, aiming to produce more natural translations with less risk of errors or hallucinations.
- Training on over seven years of proprietary data focused on content creation and translation, unlike other models trained only on public internet data.
- Input from thousands of carefully selected language experts who help train the model for high-quality translations.
DeepL says blind tests with language professionals show their translations are preferred 1.3 times more often than Google Translate, 1.7 times more often than ChatGPT-4, and 2.3 times more often than Microsoft's offering. The company reports particular improvements in English to Japanese and Simplified Chinese translations, with a 1.4-fold improvement for the combination of English and German.
DeepL CEO Jarek Kutylowski calls this launch "only the beginning" of the company's AI-powered language solutions for businesses, stating that DeepL is investing heavily in improving translation quality, efficiency, and adaptability.
The new model is available to DeepL Pro customers for translations between English, Japanese, German, and Simplified Chinese, with more languages planned. Users can activate it in the web translator by selecting "Next Gen Model."
DeepL declined to provide additional details about the model's inner workings, testing methodology, or specific performance metrics beyond what was included in their press release.
In May 2024, DeepL announced a $300 million funding round, making it one of the best-funded German AI companies alongside defense AI firm Helsing.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now