The first AI model based on Yann LeCun’s vision for more human-like AI
FacebookJun 14, 2023
Facts
- Meta Platforms announced on Tuesday that it would allow researchers access to components of a new "human-like" artificial intelligence (AI) creation model that allegedly can analyze and complete unfinished images more accurately than existing models.
- The company's AI team further stated that it is introducing the Image Joint Embedding Predictive Architecture (I-JEPA), the first AI model based on the vision of its Chief Scientist Yann LeCun to develop a new architecture to help machines learn faster, plan how to fulfill complex tasks and easily adapt to unfamiliar situations.
- Contrasting to traditional generative models that are trained on extensive datasets, I-JEPA aims to replicate human perception and common-sense reasoning by reducing visual images to abstract representations and using predictive learning.
- This comes as Meta's executives have downplayed warnings from others in the industry about the potential risks of the AI, refusing to sign a statement in May backed by top executives from OpenAI, DeepMind, Microsoft, and Google.
- Over the weekend, Meta released via Github its new music-generating tool, MusicGen, which uses AI to turn text descriptions into audio recordings after being trained using 20K hours of music. A demo version is available to try via [the AI chat model community site] Hugging Face.
- The company has made significant breakthroughs since establishing its AI research lab in 2013, open-sourcing over 30 AI models and frameworks — including the leading AI programming language PyTorch.
Spin
Pro-establishment narrative
The introduction of I-JEPA is a landmark in the development of AI, revealing the potential of self-supervised learning architectures to overcome key limitations of state-of-the-art systems. Hopefully, this approach will prove to be extendable to other domains, including video understanding and image-text paired data.
Establishment-critical narrative
Though a lot of people are looking forward to the disruptive impact of AI on society, it's undeniable that scammers and other malevolent actors are willing to exploit AI tools — including apparently harmless image generators — for unscrupulous purposes. Government regulators and cybersecurity experts must find a way to address such threats while protecting innovation and the digital freedoms of ordinary people.
Nerd narrative
There's a 4% chance that human brain emulation will be the first successful route to human-level digital intelligence, according to the Metaculus prediction community.
