Meta, Facebook's parent company, honored the 10th anniversary of its Fundamental AI Research team, emphasizing that super intelligence is a long way to go.
Meta Believes AI Models Still Far From Super Intelligence
During the event, Meta's chief scientist and deep learning pioneer Yann LeCun stated that he believes that the current AI models have still a long way to improve. He also highlighted that the technology is still decades away from developing sentience and common sense in AI models.
LeCun's statement comes in contrast with Nvidia CEO Jensen Huang who stated previously that AI will be "fairly competitive" with humans in just about five years. On the other hand, LeCun claimed that Nvidia is hopping on the AI craze as the chip company has so much to gain from it.
Most importantly, the chief scientist stated that the current focus of the tech industry is to give services based on text data, which is not enough to create the advanced human-like AI models that we imagine.
AI Needs Improvement in Training to Advance
According to LeCun, relying on text for training is a poor source of information, which explains his belief that society will be getting a "cat-level" or "dog-level" AI for years before reaching the ambitious human-level AI.
"Train a system on the equivalent of 20,000 years of reading material, and they still don't understand that if A is the same as B, then B is the same as A." He also shared no training is enough to learn a lot of basic things in the world.
Currently, the Meta AI executives are working on creating apps that could work on a wide range of data from audio, image, and video information. Other AI companies have also started to support various file formats to train their models.
Related Article : Meta Receives Complaints for 'Very Expensive' Ad-Free Fees for EU Users
