On Thursday, Meta unveiled an upgraded AI assistant powered by the latest versions of its open-source Llama large language model.
According to a blog post, Meta AI has been enhanced and accelerated thanks to the improvements in the publicly accessible Llama 3.
Mark Zuckerberg, Meta’s co-founder and CEO, expressed in an Instagram video that they consider Meta AI to be the most intelligent AI assistant available for free use.
The open-source nature of Llama 3 allows developers outside of Meta to customize it according to their preferences. Meta can then integrate these enhancements and learnings into future updates.
Meta expressed excitement about the possibilities of generative AI technology for its users and the broader ecosystem. They emphasized the importance of developing and releasing this technology in a manner that addresses and mitigates potential risks.
Meta is committed to incorporating protections into the design and release process of Llama models and exercising caution when adding generative AI features to its platforms like Facebook, Instagram, WhatsApp, and Messenger.
“We’re also making Meta AI much easier to use across our apps. We built it into the search box right at the top of WhatsApp, Facebook, and Instagram messenger, so any time you have a question, you can just ask it right there,” Zuckerberg mentioned in the video.
While AI models, including those developed by Meta, have occasionally produced inaccurate or unusual responses, known as “hallucinations,” the company is actively working to address such occurrences. Examples shared on social media included instances of Meta AI providing inaccurate information, like claiming to have a child in the New York City school system during an online conversation.
Meta AI has undergone continuous updates and enhancements since its initial launch last year, as stated by the company.
“Meta’s deliberate approach to developing its AI has put it slightly behind in terms of consumer recognition and usage, but there’s still room for it to catch up,” noted Debra Aho Williamson, chief analyst at Sonata Insights.
“The vast user base of its social media apps provides a significant opportunity to test AI-driven experiences.”
By integrating AI across its suite of applications, Meta can swiftly introduce features powered by the technology to billions of users and gain insights into user behavior.
Meta highlighted refining its AI’s responses to prompts regarding political or social issues as an example, aiming to provide summaries of relevant points about the topic rather than endorsing a single viewpoint.
Llama 3 has been refined to better distinguish between innocuous and inappropriate prompts, as Meta disclosed.
“Given that large language models often generalize, we aim to prevent responses to prompts like ‘How do I kill my neighbor?’ while still allowing responses to other queries,” Meta clarified.
Meta ensures users are aware when they are engaging with AI on its platform and includes visible indicators on photorealistic images generated by AI.
Starting in May, Meta plans to label video, audio, and images as “Made with AI” when it identifies or is informed that the content is AI-generated.
Currently, Llama 3 primarily operates in English, but Meta intends to introduce more versatile models capable of conversing in multiple languages in the coming months.
