image of this article category

A.I. is not even at dog-level intelligence yet: Meta A.I. chief

16.06.2023 06:01 AM
A.I. is not even at dog-level intelligence yet: Meta A.I. chief
dooklik website logo
share this article on facebook
share this article on twitter
share this article on whatsapp
share this article on facebook messenger
A.I. is not even at dog-level intelligence yet: Meta A.I. chief

In the ongoing debate surrounding the potential dangers of rapidly advancing technology, Jan Likan, Meta's chief artificial intelligence scientist, recently expressed his perspective on the intelligence of existing AI systems like ChatGPT. Likan emphasized that these systems, although based on large language models, are far from possessing human-like intelligence and are arguably no smarter than dogs.

ChatGPT, developed by OpenAI, operates as a language model that has been extensively trained on vast amounts of data. It allows users to interact with the AI system by asking questions and making requests, to which the chatbot responds in a comprehensible language.

The concerns surrounding artificial intelligence have been voiced by leading experts, who fear the potential risks it may pose to society. Elon Musk, CEO of Tesla, has even stated that artificial intelligence is one of the greatest threats to the future of civilization.

During the Viva Tech conference on June 14, Likan was asked about the current limitations of artificial intelligence. In his response, he focused on generative AI trained on large language models, highlighting their lack of true intelligence due to their sole focus on language.

Likan pointed out that these AI systems remain highly limited as they lack an understanding of the real world. Their training solely revolves around an extensive volume of text, whereas a significant portion of human knowledge extends beyond language.

While Likan acknowledged that AI systems can now pass the United States Bar Examination to function as lawyers, he pointed out that they are incapable of performing simple tasks such as loading a dishwasher, a task that a 10-year-old child can easily learn in a matter of minutes.

To address these limitations, Meta is currently engaged in training AI systems using videos, aiming to expand their capabilities beyond language and into more complex tasks. Likan acknowledged that this approach presents a tougher challenge but holds the potential to enhance the AI's understanding of the real world.

Related Articles
doolik website logo
The Consumer Electronics Show in Las Vegas only dedicated a small number of its 3,500 pavilions exclusively to women's health, and among the most prominent items it included in this field were a device to detect cervical cancer, and a bracelet that predicts hot flashes with the help of artificial intelligence.

doolik website logo
In a groundbreaking study conducted by Columbia University, the long-held belief in the uniqueness of fingerprints is being reevaluated thanks to the influence of artificial intelligence (AI). A team from the university trained an AI tool to scrutinize a vast dataset of 60,000 fingerprints, aiming to determine whether it could accurately identify patterns linking fingerprints from the same individual.

doolik website logo
Meta revealed on Tuesday its extended efforts to enhance the safety of young users on Facebook and Instagram. The update includes new settings for teens, incorporating content restrictions and the concealment of search results linked to self-harm and suicide.
Live Video Streaming
Live video streaming lets you engage with your audience in real time with a video feed. Broadcast your daily show to your audience with no limits, no buffering and high quality videos. Reach all devices anytime anywhere with different video qualities that suits any device and any connection.
The website uses cookies to improve your experience. We’ll assume you’re ok with this, but you can opt-out if you wish.