Orange to Develop Custom AI Models for Africa

French telecommunications giant Orange has partnered with OpenAI and Meta to create custom AI models designed to better understand regional African languages, which are challenging for most conversational systems. This was reported by CNBC.

Currently, major startups train algorithms with information from the United States, which can cause models to lose important context from other regions, such as culture and language. Such neural networks find it harder to understand information from less commonly used dialects.

“With an open model, you can perform what is called fine-tuning by introducing additional information that was not included during the initial training. We are adding West African regional languages that are not understood by any AI,” commented Steve Jarrett, head of AI at Orange.

The company will begin deploying AI models for two languages—Wolof and Fula—at the beginning of 2025. These languages are spoken by approximately 16 million people. Ultimately, the plan is to cover 18 countries in West Africa.

In addition to using artificial intelligence to improve customer service processes, the firm employs the technology to enhance the quality of mobile networks. Neural networks help identify and predict service disruptions before they are noticed by users.

AI Support from Meesho

Softbank-backed online shopping site Meesho has launched the “first AI-powered voice bot” for Indian firms operating in the e-commerce sector. The solution improves customer interactions, reducing some costs by 75%.

The startup reported results: artificial intelligence handles 60,000 user calls daily in English and Hindi. Plans are in place to add six other Indian languages.

Meesho has 160 million customers in India, 80% of whom are in smaller cities, towns, and villages.

“We receive a ton of support calls,” noted company co-founder Sanjeev Barnwal.

Instead of developing its own large language model, Meesho integrated existing solutions with its tools to understand local context and language nuances. As a result, 95% of support requests are resolved autonomously, while the remaining 5% require human intervention.

Back in November, Salesforce CEO Marc Benioff stated that the future of artificial intelligence lies with autonomous agents rather than large language models.