Musk claims Grok 5.0 will achieve AGI

Elon Musk and his AI model Grok claim that AGI will be achieved in a few iterations. Grok 5.0 “hits AGI while others are still guessing,” potentially bringing us into the era of artificial general intelligence.

Musk’s artificial intelligence (AI) company, xAI, is planning to release a new version of its AI model, Grok.

Grok 5 is due to be released between May and June, in quarter 3 of this year, and users are already anticipating its capabilities.

One X user posted a meme that represents Grok’s progression. The post received over 4 million views in just nine hours.

The post shows a hypothetical timeline of Grok versions, showing that 4.9 will be trained on 20 trillion tokens (20T) and that by Grok 5.0, we’ll have reached artificial general intelligence (AGI).

AGI is a hypothetical timeline in machine learning and AI advancement in which models will reach or exceed human intelligence.

The timeline claims that as early as Grok 6.0, the AI model will have reached artificial super intelligence (ASI), a futuristic form of AI in which models will exceed human intelligence in every way.

The billionaire CEO responded to the timeline, calling it a “banger.”

While the post is not factual in any sense, the user, who appears to be a big supporter of all Musk’s ventures, shared the fictional trajectory to their almost 200,000 followers.

Min Choi, an AI commenter, previously posted that Musk had also mapped out xAI’s path towards AGI.

The next version of Grok (4.4) will be twice the size as it's trained on one trillion parameters, Musk claims. This version will “probably be ready for release in early May.”

xAI’s plans for Grok 4.5 are likely 1.5 trillion and will be released “hopefully” by late May, the world’s richest man said via X.

One user, who is followed by Musk and claims to help the tech mogul “make X a force for good,” asked Elon directly, “Do you think we will have achieved AGI with one of these models?”

Musk simply replied, “Grok 5.0.”

Is the Grok timeline accurate?

The current version of Grok claims the meme is “spot on,” as the unreleased version (5.0) “hits AGI while others are still guessing.”

Grok claims that while other AI companies like OpenAI “play it safe” with their models, xAI is building models for “maximum truth” and is based on “zero fluff.”

By this logic, xAI’s risk-taking and lack of “fluff” means that Grok is closer to AGI than any other model in the world.

Which is patently false.

Tech leaders believe AGI is already here

While there are models like Anthropic’s Mythos, which is the first-ever model reportedly trained on 10 trillion parameters, and are among the most powerful models out there (so powerful that it’s not available to the public), this doesn’t mean that AGI has been achieved.

More parameters don’t necessarily translate to AGI.

At the start of this month, Silicon Valley venture capitalist Marc Andreessen announced the arrival of AGI, with many X users calling his bluff.

At this point, there is no agreement among academics and scientists that suggests AGI has been achieved or that it could even be achieved at all.

However, CEOs of AI technology companies repeatedly claimed that superintelligence is either here already or just around the corner.

The conversation on this topic is live. Join in the discussion.

Last year, OpenAI chief Sam Altman said, “We’re now confident we know how to build AGI,” and that this will likely happen in just a few years.

OpenAI staff member and MIT Professor Alexander Madry claims that "The scientific breakthroughs needed for AGI have already been achieved" and that we “might be able to declare AGI…by the end of 2026.”

AI models don’t have what it takes

While there’s still a lot of hype around AI, and some of it for good reason, these models are demonstrating what is called “jagged intelligence.”

At the moment, AI models learn from tonnes of data, but don’t learn from the real world, which, if AI giants want to achieve AGI, is a fundamental issue.

Most researchers tend to believe that in order to reach AGI, models need to either learn from the real world or a very sophisticated simulation of it.

For this to work, IBM Fellow Francesca Rossi said that companies must “think beyond just scaling models and focus on how intelligence works.”

So, the fictional trajectory outlined by Musk and regurgitated by his followers may not be realistic, given that the AI race has become a battle over both physical and digital resources.