CEO Mark Zuckerberg talks about plans to install image generation AI on Facebook and Instagram

Meta, which operates Facebook and Instagram, is also focusing on AI-related development, and is releasing high-performance AI such asthe large-scale language model 'LLaMA' and the animation generation AI 'Animated Drawings' one after another. Newly, CEO Mark Zuckerberg reveals that he plans to install content generation AI on Facebook and Instagram.



A recommendation system using AI is already in operation on Facebook and Instagram timelines, and recommended posts selected by AI are displayed on each user's timeline. According to CEO Zuckerberg, 20% of the posts displayed on each user's timeline were selected by AI. With this AI recommendation system, it is said that Instagram has increased user stay time by 24% and improved monetization efficiency by more than 30%.

In addition, CEO Zuckerberg is planning to use content generation AI for purposes such as `` improving the chat experience of WhatsApp and Facebook Messenger '' and `` image generation tools for Facebook and Instagram ''. In addition, he says that he is considering incorporating movie generation AI into the service in the future.

Regarding the capital investment required for AI development, CEO Zuckerberg said, ``At the moment, there is no fact that we are behind in the development of AI infrastructure, and we have large-scale processing capacity that can sufficiently handle AI development. ”, dismissing the point that capital investment is delayed compared to other companies.

The main AI development results announced by Meta after 2023 are as follows.

◆Large scale language model “LLaMA”
``LLaMA (Large Language Model Meta AI)'' announced on February 24, 2023 has recorded benchmark results that exceed GPT-3 with fewer parameters than OpenAI's large-scale language model `` GPT -3''. increase. LLaMA is open to the public for free for researchers, and many AI projects using LLaMA have been announced.

Meta announces large-scale language model 'LLaMA', can operate with a single GPU while having performance comparable to GPT-3 - GIGAZINE

◆ Animation generation AI 'Animated Drawings'
'Animated Drawings' is an AI that can create animation from a single image. At the time of article creation, a demo that can be used for free is being released. You can check the result of actually generating animation using the demo in the following article.

Meta AI releases AI 'Animated Drawings' that creates slimy moving animation from one character illustration for free - GIGAZINE

◆ 'Segment Anything Model' that cuts through the subject in the image in one shot
'Segment Anything Model' is an AI that can automatically recognize the subject in the image and realize actions such as 'select only the dog in the photo'. You can check the results by using the 'Segment Anything Model' demo in the article below.

Publish AI model 'Segment Anything Model' that allows Meta to separate and select objects in the photo - GIGAZINE

◆ Language model 'Toolformer' that can call external applications and outsource processing
Existing chat AI specializes in actions such as ``responding in natural language'', and is not good at actions such as ``solving math problems'' and ``searching for the latest information on the Internet''. 'Toolformer' developed by Meta can efficiently perform processing that language models are not good at by calling external APIs for calculation processing and searching for the latest information.

Meta's language model 'Toolformer' calls and uses search engines, calculators, calendars, etc. with API - GIGAZINE