OpenAI and Meta are Working on Creating AGI

OpenAI and Meta are preparing to introduce new versions of their large-scale AI language models that will gain the ability to reason and plan actions. Meta promises to present Llama 3 in the next few weeks. The release of GPT-5 from OpenAI is expected this summer. They will probably be the first confident step towards the creation of artificial general intelligence (AGI) and the achievement of superhuman intelligence by machines.

Meta Vice President Joelle Pineau and OpenAI Chief Operating Officer Brad Lightcap emphasized that the new models will not only be able to conduct a dialogue but also reason, plan, remember, and predict. In addition, Brad Lightcap stated that the next GPT model will be much better at solving complex cognitive problems.

At a recent Meta AI Day event in London, Meta's chief artificial intelligence scientist Yann LeCun noted that existing AI models process information sequentially but without planning, which often leads to errors. Improving reasoning abilities will allow AI to analyze options and plan its actions, anticipating their consequences. This will significantly increase the level of intelligence of machines.

Moreover, LeCun spoke about the development of AI agents based on a new model that will independently plan complex tasks. Meta intends to implement them in WhatsApp and Ray-Ban smart glasses, adapting Llama 3 for various applications. For example, cameras in Ray-Ban Meta glasses could be used to inspect a broken coffee machine. An AI assistant integrated into them, powered by Llama 3, will explain to the owner how to fix the breakdown. LeCun envisioned a future where AI assistants would become an integral part of our digital communications, radically changing our digital experiences.

Today, we are seeing an accelerating race to improve generative AI, with major players like Meta, OpenAI, and Google launching new models this year. Their achievements are critical to the development of what researchers call artificial general intelligence, which mimics human thinking and cognition.