ChatGpt and other AI models will not reach human intelligence, for now. Word of Meta
Meta's lead AI scientist, Yann LeCun, explained to the Financial Times that current AI models (like ChatGpt) are imperfect. In the meantime, the social giant is working on an alternative approach to create "superintelligence" in machines
The AI model that powers systems like OpenAI's ChatGpt and Google's Gemini will not be able to reach human intelligence levels.
This is what Meta's head of artificial intelligence, Yann LeCun, considered one of the founding fathers of AI, claims in an interview with the Financial Times .
The head of AI at the social giant that owns Facebook and Instagram told the FT that the large language models that power generative artificial intelligence products like ChatGpt will never achieve the ability to reason and plan like humans. The scientist argued that these models are based on pre-existing training data and are therefore “inherently dangerous” as they can only provide accurate answers as part of their training.
Meta has invested heavily in Large Language Models (Llms), i.e. large language models, to keep up with competitors such as OpenAI and Google.
So much so that Meta founder Mark Zuckerberg announced the launch of the new generation of the Meta Llama 3 language model available in open source mode “later this year” along with greater integration of the Meta AI virtual assistant with WhatsApp, Instagram, Facebook and Messenger.
But it doesn't end here. Meta is focusing on a radical alternative approach to creating “superintelligence” in machines. The company founded by Mark Zuckerberg is working to develop an entirely new generation of artificial intelligence systems that it hopes will power machines with human-level intelligence.
Although LeCun admitted to the Financial Times that this vision could take 10 years to achieve.
All the details.
CHATGPT'S AI MODEL WILL NOT REACH HUMAN INTELLIGENCE
LeCun told the Financial Times that large language models, the kind of AI that drives popular generative AI tools like OpenAI's ChatGPT, Meta's Llama and Google's Gemini, despite their impressive performance on certain tasks, are fundamentally limited and will never achieve the ability to reason and plan like human beings.
He pointed out their shortcomings, citing their lack of understanding of logic, limited understanding of the physical world, lack of persistent memory, inability to reason, and inability to hierarchical plan.
NOT EVEN THE MOST ADVANCED VERSIONS
His comments come as Meta and its rivals push forward with more advanced LLMs. As the FT recalls, OpenAI head Sam Altman believes it provides a fundamental step towards the creation of artificial general intelligence (AGI), the point at which machines will have greater cognitive capabilities than humans.
OpenAI last week released its new, faster GPT-4o model, and Google unveiled a new “multimodal” AI agent capable of answering real-time queries on video, audio and text called Project Astra, powered by an updated version of his Gemini model.
Meta also launched its new Llama 3 model last month. The company's head of global affairs, Sir Nick Clegg, said its latest LLM has "significantly improved skills such as reasoning", i.e. the ability to apply logic to queries.
However, LeCun said this evolution of LLMs has been superficial and limited, with the models learning only when human engineers step in to train them on that information, rather than the AI arriving at a conclusion as organically as people .
META WORK ON “SUPERINTELLIGENCE”
Meanwhile, the Menlo Park group's AI chief leads a team of about 500 employees at Meta's Fundamental AI Research (Fair) lab, where they are working to create an AI that can develop common sense and learn how it works the world similarly to humans, in an approach known as “world modeling.”
But it could take up to 10 years to achieve human-level artificial intelligence using the world modeling approach, LeCun predicted.
ZUCKERBERG'S STRATEGY
But will Meta investors be so patient?
Last month, Meta burned through nearly $200 billion of its market value after releasing first-quarter results, posting its biggest decline since October 2022
As the accounts were released, CEO Mark Zuckerberg vowed to increase spending and transform the social media group into "the leading artificial intelligence company in the world", spooking Wall Street investors worried about rising costs with little immediate revenue potential.
“In recent months, Zuckerberg has made artificial intelligence a priority, refocusing Meta on the technology after OpenAI released its ChatGpt chatbot in 2022, sparking a frenzy of competition and development among big tech makers,” Bloomberg recently noted .
And he also maintained this course in the presentation of the quarterly . During the call with investors, CEO Mark Zuckerberg focused most of his comments on artificial intelligence. He said Meta wants to be "the world's leading AI company" and would have to increase spending "significantly before we get a lot of revenue from some of these new products."
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/chatgpt-e-gli-altri-modelli-di-ia-non-raggiungeranno-lintelligenza-umana-per-ora-parola-di-meta/ on Sun, 02 Jun 2024 14:10:58 +0000.