Article complete

Meta AI Releases A Multimodal Model "CM3leon" — But Won’t Release It

Meta AI's model has the potential to revolutionise the approach in which current users interact with and manipulate visual content.

A small cactus in a pot, wearing a straw hat and neon sunglasses, set against a sandy desert backdrop.

Key takeaways

  • Meta AI has introduced CM3leon, a multimodal model for text-to-image and image-to-text generation.
  • The model significantly improves the interaction between users and visual content.
  • CM3leon requires less computing power and a smaller training dataset than previous methods.
  • It employs a Transformer architecture, distinguishing itself from the diffusion method commonly used in image generation.
  • Despite its advanced capabilities, CM3leon remains closed source, frustrating AI enthusiasts.

Takeaways drafted with AI from the article text.

The AI model bridges the gap between text and images. With its capabilities in text-guided image generation and editing. The model  has the potential to revolutionise the approach in which current users interact with and manipulate visual content.

According to Meta, its model boasts improved capabilities in producing coherent imagery that closely aligns with input prompts. What sets CM3leon apart is its efficiency, as it requires only five times the computing power and a smaller training dataset compared to previous transformer-based methods.

Meta touts CM3leon's prowess in various vision-language tasks, including visual question answering and long-form captioning. This novel approach by Meta marks a departure from the diffusion method commonly employed in image generation. Instead, the company's researchers opted for the Transformer architecture, a neural network design widely recognized for its successful implementation in large language models like OpenAI's GPT-4. While CM3leon is not the first transformer-based image generator—StyleSwin precedes it—Meta asserts that it surpasses other contenders in terms of efficiency.

Though the model is being lauded for being state-of-the-art the fact that it is closed source makes the research community irk. Similarly, last month the tech giant unveiled Voicebox but did not make the model public over fears of potential misuse. “While we believe it is important to be open with the AI community and to share our research to advance the state of the art in AI, it’s also necessary to strike the right balance between openness with responsibility,” Meta had stated while releasing the paper.  

Again, Meta is not the only tech goliath gatekeeping its research. Even though companies like Google, Microsoft, Facebook, Amazon, and Apple have contributed tons to research over the years — they also have kept a major chunk of it to themselves. The reason this displeases the research community is because these companies have been dependent on the open source community yet refuse to contribute as much as they gain. 

Read more: Big Techs Flip-Flop on Open Source

Published Updated

Spotted an error? Tell our editors. Read our editorial standards and corrections.

More in this story

Meta · 5 stories
  1. Meta gives away its language model for free
  2. You are here: Meta AI Releases A Multimodal Model "CM3leon" — But Won’t Release It
  3. How Meta Broke the Barriers To Entry With LLaMA 2
  4. Llama 3 Might Not be Open Source
  5. Meta Reports 33% Revenue Growth, Testing Early Versions of Business AI
About the author

Contributor

All stories
Belamy · Daily

The day’s AI news, before your first meeting.