Upcoming OpenAI, Meta models to have ‘reasoning’ capabilities: A step closer to AGI?

As Meta’s VP of AI Research, Joelle Pineau, puts it: “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.”

On the OpenAI side, COO Brad Lightcap hinted to the publication that their upcoming release, likely dubbed GPT-5, will make meaningful strides on these “hard problems” of reasoning. Lightcap acknowledged that current AI is “really good at one-off small tasks” but still quite narrow in scope.

The plan is for these new AI assistants to be capable of complex, multi-stage operations. Yann LeCun, Meta’s chief AI scientist, envisions AI “agents” that could plan an entire travel itinerary, booking all the requisite flights, hotels, and transportation from start to finish.

LeCun describes current language models as essentially “producing one word after the other really without thinking and planning.” The enhanced reasoning should allow the AI to “search over possible answers,” “plan sequences of actions,” and model out the effects before execution.

Across the industry, technology leaders like Google and Anthropic are also working feverishly on their own upgraded large language models, according to the report.

Both OpenAI and Meta are expected to start deploying their new reasoning-enabled models in the coming weeks and months. OpenAI’s Lightcap teased “more to say soon” on GPT-5’s release, while Meta is preparing to launch Llama 3 in various sizes optimised for different apps and devices, including WhatsApp and their Ray-Ban smart glasses.

There may be no fixed definition of AGI but companies are in a race to achieve it nevertheless, trying to overcome technical hurdles around replicating aspects of human cognition like reasoning, abstraction, and contextual understanding in machines. Researchers view capabilities like multi-step planning and open-ended reasoning as crucial milestones.