The list of flagship AI models missing their promised launch dates continues to grow.
Last summer, Elon Musk, founder and CEO of AI company xAI, announced that Grok 3, xAI’s next major AI model, would be released by the “end of year” 2024. Grok, which competes with models like OpenAI’s GPT-4 and Google’s Gemini, can analyze images and respond to queries, and powers various features on X, Musk’s social platform.
Grok[.]com is possibly coming soon with Grok 2.5 model (grok-2-latest – “Our most intelligent model”) – thanks for the hint, anon! pic.twitter.com/emsvmZyaf7
— Tibor Blaho (@btibor91) December 20, 2024
xAI:
In a July post, Musk wrote, “Grok 3 end of year after training on 100k H100s should be really something special,” referring to xAI’s vast cluster of GPUs in Memphis. “Grok 3 will be a major leap forward,” he reiterated in a follow-up post in mid-December.
However, as of January 2, Grok 3 has yet to arrive, and there are no signs of its imminent release.
In fact, some code discovered on xAI’s website by AI tipster Tibor Blaho suggests an intermediate model, “Grok 2.5,” might be released first. This isn’t the first time Musk’s ambitious goals have missed their targets; his predictions for product launch timelines are often unrealistic. To be fair, during an August interview with podcaster Lex Fridman, Musk acknowledged that Grok 3 would “hopefully” be available in 2024 “if we’re lucky.”
The delay of Grok 3 is noteworthy as it fits into a broader pattern.
Last year, AI startup Anthropic also failed to deliver the successor to its top model, Claude 3 Opus. Despite announcing the release of a next-gen model, Claude 3.5 Opus, by the end of 2024, the company eventually removed all references to it from its developer documentation. (Reports suggest that Anthropic had completed training for Claude 3.5 Opus but ultimately decided against releasing it due to economic considerations.)
Similarly, Google and OpenAI have encountered setbacks with their flagship models in recent months.
This trend could point to the limitations of current AI scaling methods. In the past, substantial improvements in model performance were achieved by training them with vast amounts of computing power and data. However, the performance gains with each new generation are becoming smaller, prompting companies to explore alternative strategies. Musk himself acknowledged this challenge during his Fridman interview.
ELON MUSK:
When asked if Grok 3 would be state-of-the-art, Musk replied, “Hopefully… I mean, this is the goal. We may fail at this goal. That’s the aspiration.”
Other factors could contribute to Grok 3’s delay, such as xAI’s smaller team compared to its competitors. Regardless, the delay adds to the growing evidence that traditional AI training methods may be reaching their limits.
