Yann LeCun, Meta’s AI chief, believes that large language models like ChatGPT will never achieve human-like reasoning and planning abilities. Instead, he is working on a new approach to create “superintelligence” in machines. LeCun argues that relying on advancing LLMs to achieve human-level intelligence is unsafe and limited. Meta is investing heavily in developing its own LLMs to catch up with competitors like OpenAI and Google.
LeCun’s team at Meta’s Fair lab is focused on creating AI systems with human-level intelligence through an approach called “world modeling.” This vision is a long-term goal that could take up to 10 years to achieve. Despite the risks and costs involved, LeCun is determined to push forward with this experimental vision.
While Meta and its rivals continue to enhance LLMs, LeCun believes that these models are limited and superficial. He is exploring alternative methods to achieve artificial general intelligence, such as reinforcement learning. LeCun’s team is testing different ideas, including feeding systems with video data to mimic how children learn from observing the world.
Some experts are skeptical of LeCun’s vision, citing challenges in teaching AI models common sense and causality. However, LeCun remains focused on developing AI agents with human-level intelligence that can be interacted with through wearable technology. Despite criticisms and doubts, LeCun is determined to push forward with his vision of creating superintelligent machines.
Source link
Source link: https://www.ft.com/content/23fab126-f1d3-4add-a457-207a25730ad9
GIPHY App Key not set. Please check settings