Did Grok 3 also "miss the deadline"? Can Musk's "100,000 GPUs with the strongest computing power" not handle the "next generation AI large model"? |【Hard AI】

Source: Hard AI

Author: Jiang Zihan

Are AI industry giants delaying the release of the next generation of models indicating that AI technology development has entered a bottleneck?

Last summer, Elon Musk posted on social platform X that xAI's next generation AI model Grok 3 would be trained on 100,000 H100 GPUs and would be launched by the end of 2024, stating it "will be very special." However, it is now 2025, and Grok 3 has yet to be released, with no signs of an imminent launch.

It is noteworthy that, apart from xAI, other AI companies are experiencing similar situations:

Anthropic has canceled the planned release of the Claude 3.5 Opus model by the end of 2024;

Reports indicate that the flagship model development of Google and OpenAI has also encountered setbacks.

Analysts suggest that the relatively small size of the xAI team may be one of the reasons for the delay in the release of Grok 3.

But more importantly, the delays in releases by AI companies may reflect the current limitations of AI scaling laws—previously, using massive computing power and larger datasets to train models could significantly enhance AI model performance; however, recently, the performance improvements of each generation of models have begun to diminish, prompting tech giants to seek alternative technologies.

Musk himself hinted at this in an interview: when asked if Grok 3 would reach industry-leading levels, Musk replied, "I hope so, that’s our goal, but we might fail."