AI pioneers OpenAI and Meta are nearing the release of AI models capable of planning and reasoning. Could this be a significant leap towards artificial intelligence surpassing human cognitive abilities?
While AI has achieved remarkable progress in recent years, it still needs to catch up to human capabilities in specific domains. However, this gap might close soon. OpenAI and Meta executives anticipate releasing significantly more powerful AI models.
According to the OpenAI and Meta report, ChatGPT-5 and Llama-3 will surpass current text-generating models. These companies suggest their creations will exhibit capabilities much closer to human reasoning.
“We are hard at work in figuring out how to get these models not just to talk, but actually to reason, to plan… to have memory,” Joelle Pineau, vice president of AI research at Meta, explained.
In a statement to MailOnline, Meta announced the upcoming release of early access versions of their Llama 3 model. This rollout will be followed by a series of additional models throughout 2024. OpenAI, on the other hand, is planning to release GPT-5 “soon,” executives told the Financial Times.
OpenAI’s popular chatbot, ChatGPT, leverages large language models (LLMs) to anticipate the most likely sequence of words in a sentence. By ingesting vast amounts of data, these LLMs develop impressive proficiency in stringing words together, creating an illusion of sentience and reason.
However, Meta’s chief AI scientist, Yann LeCun, says the most advanced versions of AI can still “make stupid mistakes.” LeCun, who has previously cautioned against expecting near-human reasoning abilities in AI for the foreseeable future, explained at a Meta event in London that current AI systems operate by reactively generating text sequences without genuine thought or strategic planning.
LeCun further elaborated that equipping AI with reasoning capabilities would empower these systems to consider various potential solutions, formulate action plans, and create a “mental model of what the effect of [its] actions are going to be.”
Brad Lightcap, OpenAI’s COO, suggests that we are only beginning to explore the potential of these models to reason. “We’re going to start to see AI that can take on more complex tasks in a more sophisticated way,” the top executive said.
Lightcap suggests that as AI models take on longer and more complex tasks, they will inherently need to develop better reasoning skills, such as considering past experiences and planning for future actions.
The quest for human-level AI: What is AGI?
According to LeCun, Meta is developing advanced AI systems (he calls them “agents”) that can book and plan an entire trip, like a Paris to New York itinerary. This development could represent a substantial breakthrough in AI capabilities and a step closer to creating artificial general intelligence (AGI).
AGI remains the ultimate objective for many AI researchers. It seeks to create machine-learning systems with human-level cognitive abilities across all tasks. Achieving this would require the AI to master complex task sequences and anticipate the outcomes of its actions.
While Demis Hassabis, CEO of Google DeepMind, expressed optimism last year about achieving human-level AI in the near future, many researchers remain cautious. They share different confidence about the imminent arrival of AGI.