Meta’s Llama 3.1 Enables AI Startups Fend Off Data Quality Issues

Meta, Telangana Govt To Tap Into AI For Boosting E-Governance & Citizen Services

SUMMARY

Meta believes that Llama can generate synthetic data to better understand languages like Hindi, Kannada, or Marathi

Meta launched its Llama 3.1 on Tuesday (July 23) to take on other AI models like GPT-4.0 and Claude 3.5 Sonnet

This comes at a time when it is being seen that general-purpose large language models are struggling with language nuances

Two days after rolling out its largest-ever open source AI model Llama 3.1, social media giant Meta has said that it will be helping Indian startups with synthetic data generation support to overcome real-world quality data issues in the local language.

Meta launched its Llama 3.1 on Tuesday (July 23) with 405 Bn parameters, trained on 15 Tn tokens using powerful Nvidia GPUs. With this LLM model, the company will compete against AI models like GPT-4.0 and Claude 3.5 Sonnet.

“We think of Llama as this general base model, that companies and developers like (Indian AI startup) Sarvam should be able to customise to bring the nuances of language and culture,” Meta’s vice president of product management Ragavan Srinivasan was quoted as saying by ET.

This comes at a time when it is being seen that general-purpose large language models are struggling with language nuances. However, Meta believes that Llama can generate synthetic data to better understand languages like Hindi, Kannada, or Marathi.

On the launch of Llama 3.1, Meta in its blog post said, “We’re taking a different approach with this release. We’re building teams internally to enable as many developers and partners as possible to use Llama, and we’re actively building partnerships so that more companies in the ecosystem can offer unique functionality to their customers as well.”

The new model will include features like enabling developers to create their custom agents and new types of agentic behaviours, a reference system and new security and safety tools. Besides, the company also plans for a feedback facility on the Llama Stack API to make it easier for third-party projects leveraging Llama models.

This comes close on the heels of Meta ramping up language options for its AI chatbot across all applications, including WhatsApp.

Meta AI will now be available in seven new languages, including Hindi and Hindi-romanised script. Besides, it will also support other languages such as French, German, Italian, Portuguese and Spanish.

Currently, Meta AI is available in 22 countries, with the latest additions being Argentina, Chile, Colombia, Ecuador, Mexico, Peru and Cameroon.

Earlier this month, Meta expanded its verified service to Facebook and Instagram users after rolling out the service for WhatsApp business users.

Meta Verified is a subscription service that helps brands enhance credibility with new audiences with a “blue tick”, enhanced account support, impersonation protection, and additional features to support discovery and connection.

India is home to the largest user base of Meta’s family of apps. While Facebook accounts for 378 Mn users in the country, WhatsApp and Instagram have 478 Mn and 362 Mn users respectively, as per Statista.