Text-Based AI Won’t Result in Human-Like Intelligence, Meta Exec Says
Despite what seems like weekly advancements in the artificial intelligence field, some are still keeping their expectations measured. Yann LeCun, Meta’s VP and chief AI scientist, for example, said at a recent event that he doesn’t think AI “super intelligence” is coming anytime soon.
In fact, LeCun went so far as to say current AI systems are decades away from reaching some semblance of sentience, CNBC reports. Instead, the AI scientist says we’re more likely to see “cat-” or “dog-level” AI and that it will be years before human-level AI is attained.
Though some have become quite bullish on machine learning advances due to the numbers or text-based models debuting or advancing, LeCun points to this as a source of skepticism.
“Text is a very poor source of information,” LeCun said, according to CNBC. The outlet added that the Meta employee explained it would take 20,000 years for a human to read the amount of text used to train modern language models and pointed out that these models “still don’t understand that if A is the same as B, then B is the same as A.”
LeCun also waved away notions of AI replacing humans, pointing to the fact that AI is trained on language, using its ability to pass a bar exam but inability to load a dishwasher as a 10 year old can as an example, Business Insider reports.
Still, LeCun is not abandoning AI. He and other Meta AI executives are researching how the models used to make apps and program like OpenAI’s ChatGPT can working with other kinds of days, like audio, images, and video. ChatGPT’s recent additions that process images and voice prompts suggests OpenAI is likely already at work on such concepts.
LeCun also took time at the event to address Nvidia’s opposing stance. As CNBC notes, Nvidia CEO Jensen Huang recently said AI will be fairly competitive with humans under five years, claiming the AI would be able to best people are a multitude of mentally intensive tasks. LeCun brushed off these claims by asserting that Huang is profiting from the AI boom.
That’s isn’t necessarily untrue, either. Nvidia has been able to sell its beefy graphics units beyond its typical audience of hardcore gamers and professional creatives. CNBC points out that Meta itself used 16,000 Nvidia A1000 GPUs to train its own Llama AI software.
At the same time, LeCun has taken aim at quite a few in the AI industry. On X, formerly Twitter, the scientist accused OpenAI CEO Sam Altman, DeepMind boss Demis Hassabis, and Anthropic’s Dario Amodei of “fear-mongering” and “massive corporate lobbying” to serve their own interests, Business Insider reports. LeCun has also criticized OpenAI and Google’s DeepMind.
Image credits: Header photo licensed via Depositphotos.
