This week, Meta - parent company of Facebook - held a media event in San Francisco to mark the 10-year anniversary of their Fundamental AI Research team. Yann LeCun, Meta's chief scientist, expressed his opinion that society is likely to get AI at 'cat-level' or 'dog-level' before attaining human-level AI. The tech giant is not pouring in resources into quantum computing like Google and Microsoft.
Yann LeCun, Meta's chief scientist and deep learning pioneer, recently expressed his view that achieving AI systems with some semblance of sentience and common sense, capable of pushing beyond merely summarizing text, is still decades away. This stands in casual contrast to the opinion of Nvidia CEO Jensen Huang, who said AI would become "fairly competitive" with humans in less than five years. LeCun referred to Huang's stance by saying that there is an "AI war" and he is "supplying the weapons". Along with LeCun, other Meta AI executives have been researching how transformer models could be used with audio, image, and video data to uncover hidden correlations and potentially enable more fantastical feats than summarizing text. To this end, Meta showed a demo with a person wearing their AR glasses being able to observe visual cues when playing tennis. Such multimodal AI systems require three-dimensional visual data in addition to text and audio. As more firms like Meta and Alphabet research more advanced AI models, Nvidia's profits may rise accordingly.
Nvidia has been a key beneficiary of generative AI, as its expensive graphics processing units became the commonly accepted tool for training massive language models. Meta needed 16,000 Nvidia A100 GPUs to train its Llama AI software, prompting CNBC to query whether the tech industry will require more hardware providers as Meta and other investigators continue their work on these types of complex AI models. LeCun's response was, "It doesn't require it, but it would be nice," before noting that GPU technology is still the best choice when it comes to AI. He then went on to say that the computer chips of the future may not be GPU's, but rather neural deep learning accelerators.
When it comes to quantum computing, LeCun expressed some reservations. Many researchers outside of Meta believe quantum computing machines could be used to increase advancements in data-intensive fields such as drug discovery because they can conduct several calculations using quantum instead of conventional binary bits. LeCun, however, had his doubts and asserted that the number of problems that can be solved with quantum computing can be done much more efficiently with traditional computers. He also noted that quantum computing is a fascinating scientific topic, but it is yet to be determined if useful quantum machines will eventually become available.
Meta's Senior Fellow and former tech chief Mike Schroepfer agreed, and said he assesses quantum technology every few years, though he believes useful quantum machines may arrive eventually - just not anytime soon. He commented that when they began their AI lab a decade ago, it was clear that this technology would be ready for commercialization within the next few years.
Comments