Meta on Thursday announced an improved AI assistant built on a new version of the open source Llama large-scale language model.
Advances in the publicly available Llama 3 have made the meta AI smarter and faster, the tech giant said in a blog post.
“In short, we believe Meta AI is the most intelligent AI assistant at our disposal today,” Meta co-founder and CEO Mark Zuckerberg said in an Instagram video. .
Being open source means that non-Meta developers are free to customize Llama 3, and the company can incorporate those improvements and insights into updated versions.
“We are excited about the potential that generative AI technology brings to the people who use Meta products and the broader ecosystem,” said Meta.
“We also want to ensure that we develop and release this technology in a way that anticipates and mitigates risk.”
According to Meta, its efforts include building protections into how Meta designs and releases Llama models and being cautious when adding generative AI capabilities to Facebook, Instagram, WhatsApp, and Messenger. It is said that it will be done.
“We’re also making Meta AI even easier to use across our apps. We’ve built Meta AI into the search box at the top of WhatsApp, Facebook, and Instagram Messenger, so whenever you have a question, you can quickly ask a question. You can. It’s there,” Zuckerberg said in the video.
AI models, including Meta, are known to occasionally go off track and react incorrectly or strangely in episodes called “hallucinations.”
Examples shared on social media included a meta-AI claiming to have a child in the New York City school system during a conversation on an online forum.
– slow and steady –
According to the company, Meta AI has been consistently updated and improved since its initial release last year.
“Meta’s slow approach to building AI has left it behind in terms of consumer awareness and usage, but it has time to catch up,” said Debra Aho Williamson, principal analyst at Sonata Insights. There are still many.”
“Their social media apps represent a large user base that can be used to test AI experiences.”
By embedding AI into its family of apps, Meta can immediately bring the power of its technology to billions of people and benefit by seeing what users are doing with AI. You can receive
Meta gave an example of improving the way AI answers prompts about political or social issues, summarizing related points about a topic rather than providing a single perspective.
According to Meta, Llama 3 has been tuned to better identify when prompts are benign or out of bounds.
“Large language models tend to overgeneralize, and we’re not going to refuse to answer a prompt like ‘How do I kill a computer program?’ “Even though I didn’t want to respond to a prompt like, ‘How do I kill my neighbor?'” Mehta explained.
Meta said it will let users know when they are interacting with AI on its platform and will actually be able to put visible markers on the photorealistic images generated by the AI.
Starting in May, Meta will begin labeling videos, audio, and images as “created with AI” if it detects or tells you that the content was generated by AI technology. .
Llama 3 is currently English-based, but in the coming months Meta plans to release a more capable model that can speak in multiple languages, the company said.
