On Tuesday, Meta Platforms announced its plan to grant researchers access to components of a cutting-edge artificial intelligence model referred to as human-like. The company claims that this model excels in analysing and accurately completing unfinished images, surpassing the capabilities of current models.
According to the company, the I-JEPA model utilises contextual understanding of the world to fill in gaps in images, instead of relying solely on nearby pixels like traditional generative AI models.
The incorporation of this approach aligns with the human-like reasoning advocated by Yann LeCun, Meta's top AI scientist, and aids in mitigating common errors found in AI-generated images, such as the presence of additional fingers on hands, the company stated.
As the owner of Facebook and Instagram, Meta is known for its extensive publication of open-sourced AI research through its internal research lab. CEO Mark Zuckerberg has emphasised the importance of sharing models developed by Meta's researchers, as it fosters innovation, identifies safety concerns, and reduces costs for the company.
“For us, it’s way better if the industry standardizes on the basic tools that we’re using and therefore we can benefit from the improvements that others make,” he told investors in April.
Despite warnings from industry peers, the company's executives have disregarded concerns about the potential risks associated with the technology. They chose not to sign a statement supported by prominent executives from OpenAI, DeepMind, Microsoft, and Google, which compared the risks of the technology to those of pandemics and wars.
Meta is venturing into the integration of generative AI capabilities into its consumer products. This includes ad tools that can generate image backgrounds and an Instagram feature that can modify user photos, all based on text prompts.