The list of flagship AI models that missed their promised launch window continues to grow.
Last summer, billionaire Elon Musk, founder and CEO of AI company xAI, said Grok 3, xAI's next big AI model, would arrive by “the top of the yr” in 2024. Grok, xAI's answer to models like OpenAI's GPT-4o and Google Gemini, can analyze images and answer questions and supports a variety of features on X, Musk's social network.
“Grok 3 at the top of the yr after training for 100,000 H100s ought to be really special,” Musk said wrote in a July post on X related to xAIs huge cluster based in Memphis of GPUs. “Grok 3 can be an enormous step forward,” he said said in a follow-up post in mid-December.
But it's January 2nd and Grok 3 hasn't arrived yet – nor are there any signs that its launch is imminent.
In fact, a code on xAI's website was discovered by AI tipster Tibor Blaho suggests that an intermediate model, “Grok 2.5,” could land first.
Grok(.)com could also be coming soon with the Grok 2.5 model (grok-2-latest – “Our smartest model”) – thanks for the tip, Anon! pic.twitter.com/emsvmZyaf7
— Tibor Blaho (@btibor91) December 20, 2024
Admittedly, this isn't the primary time Musk has set a lofty goal and missed it. It is well established that Musk's statements concerning the timing of product launches are sometimes unrealistic at best.
And to be fair, in an interview with podcaster Lex Fridman in August, Musk said said that Grok 3 would “hopefully” be available in 2024, “if we’re lucky.”
But Grok 3's MIA status is interesting since it's a part of a growing trend.
Last yr, AI startup Anthropic didn’t deliver a successor to its top model Claude 3 Opus. Months after Anthropic announced that a next-generation model, Claude 3.5 Opus, would arrive by the top of 2024 scrapped All mentions of the model could be present in the developer documentation. (After According to a report, Anthropic accomplished training on Claude 3.5 Opus sometime last yr, but decided that releasing it didn't make industrial sense.)
Google and OpenAI have reportedly done this too suffered setbacks with their flagship models in recent months.
This might be evidence of the boundaries of current AI scaling laws – the methods firms use to extend the capabilities of their models. Not way back, it was possible to realize significant performance gains by training models with enormous computing power and ever larger data sets. But with each generation of models, profits shrink, causing firms to change to alternative technologies.
Grok 3 trains with 10 times, soon to be 20 times, the computing power of Grok 2
— Elon Musk (@elonmusk) September 21, 2024
Musk himself pointed this out within the Fridman interview.
“You hope (Grok 3) is state-of-the-art?” Fridman asked.
“Hopefully,” Musk replied. “I mean, that’s the goal. We could fail at this goal. That is the claim.”
There might be other reasons for Grok 3's delay. For one thing, xAI has a much smaller team than lots of its competitors. Nevertheless, the postponed launch timeframe is further evidence that traditional AI training approaches are reaching their limits.
