The Quest for Superintelligence: Meta’s Ambitious Plan
In an interview with the Financial Times, Jan LeKun, Meta’s top AI scientist, revealed his ambitious plan to develop an AI model that will surpass the capabilities of Open AI’s GPT by a wide margin. LeKun, one of the world’s four largest artificial intelligence scholars, believes that large language models (LLMs) like ChatGPT will never be able to think and plan like humans.
“LLM has a very limited understanding of logic, it doesn’t understand the physical world, it doesn’t have lasting memories, it can’t be inferred by any reasonable definition, it can’t plan hierarchically,” LeKun explained. He cited the example of a ball falling and bouncing, a simple concept that humans understand instinctively, but LLMs cannot comprehend.
Illustration of AI model
LeKun’s solution is to develop a new generation of AI systems that can run machines with human-level intelligence. This ambitious project, which could take up to 10 years to realize, aims to create a radical new approach to AI development.
The Limitations of LLMs
LeKun emphasized that LLMs are inherently unsafe because they can only provide accurate answers when the correct training data is entered. This limitation is due to their lack of understanding of the physical world and their inability to think and plan like humans.
Illustration of AI limitations
Meta’s Approach
Meta is pouring billions of dollars into developing its LLM and is chasing rival tech groups such as Microsoft-backed OpenAI and Alphabet Inc’s Google. LeKun leads a team of about 500 employees at the AI lab, where they are testing various ideas to achieve human-level intelligence.
One of the approaches being explored is video inference, which involves providing hours of video to AI systems, deliberately omitting frames, and then making the AI predict what will happen next. The institute is also studying how to build a universal text encoding system and apply it to video and audio.
Illustration of video inference
The Future of AI
LeKun’s vision for the future of AI is one where machines can think and plan like humans. While this may seem like science fiction, LeKun is confident that it can be achieved within the next decade. With Meta’s vast resources and LeKun’s expertise, it will be exciting to see how this ambitious project unfolds.
Illustration of AI future
