European AI startups race to improve chatbot language skills

Applications of AI

European startups are racing to solve the problem of popular artificial intelligence chatbots: the quality of responses in languages ​​other than English.

Helsinki-based Silo AI announced Wednesday that it will develop new large-scale language models in European languages, including Swedish, Icelandic, Norwegian and Danish, which will serve as the basis for generative AI products such as OpenAI’s ChatGPT and Google’s Bard. We plan to launch an initiative to help build it.

The Finnish company has joined other groups working to improve the technology behind chatbots that provide realistic answers to prompts written in languages ​​such as German, Hebrew and Arabic. I am joining.

The move comes as companies around the world began to adopt AI software built by the likes of Microsoft-backed OpenAI and Google. Critics have raised concerns about the overreliance on technology.

“European initiatives . increase.

Google Bard currently only works in English. His ChatGPT on OpenAI supports dozens of languages, including European languages, Hindi, Farsi, and more. However, according to those who have tested it extensively, it is not equally accurate in all languages.

Silo AI is trying to solve this problem by assembling a team of experienced AI scholars from all over Europe. They build, train, and operate their Scandinavian models on LUMI, the most powerful supercomputer on the continent, in Finland, modified to run generative AI software.

The new team initiative, known as SiloGen, plans to expand to more languages ​​over time.

However, the issue is not purely a language issue. Building the model in Europe ensures that the quality of the data used for training is representative of the culture and ethics of countries outside the United States, including privacy concerns, Sarlin said.

Silo AI models will also be open source. That means it can be analyzed and adapted by anyone who wants to deploy it. This is in contrast to OpenAI and Google’s closed models. In a closed model, companies are reluctant to share sensitive or proprietary data.

Other European efforts include OpenGPT-X and LEAM, German-led initiatives to develop open source language models. The OpenGPT-X model is built in collaboration with German AI startup Aleph Alpha.

At its launch last year, the group behind OpenGPT-X warned that the lack of access to model details such as GPT-4 threatened Europe’s “digital sovereignty and market independence” in AI, prompting European firms and He warned that it could hinder research growth.

Marco Trombetti, chief executive of Italian digital translation company Translated, said the leading chatbots were programmed to give the best results in English, which “is not fair to other countries”. said.

To combat this, his company created ChatGPT’s live translation tool. The tool works in 60 languages ​​and aims to improve the tool’s answers.

Regarding current generative AI tools, Trombetti said:

These concerns are not limited to Europe. The Israel Innovation Authority said he spent 7.5 million shak (about $2.1 million) to found the Natural Language Processing Association. The group seeks to reverse “the poor quality and inadequacy of Hebrew and Arabic speech recognition in various kinds of computerized systems,” said its chief executive, Dror Bin. says Mr.

Binh said funding for AI research is limited in Arabic-speaking countries and relatively few Hebrew-speaking countries in the world lead to the integration of AI products into commercial applications such as Microsoft Office and Google Workspace. I fear that I will be left behind as time goes on.

“The quality of understanding and recognizing human speech in Hebrew and Arabic is poor, which is a barrier to achieving and applying advanced services,” he added.

Additional reporting by John Thornhill

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *