CEO Elon Musk officially announced the launch of the bogus intelligence (AI) chatbot ‘Grock 3’. Attention is concentrated on how the model using the world’s largest GPU will perform.
CEO Musk announced on the sixth (local time) through No specific time was mentioned.
This appears to be a response to TechCrunch’s report last week that “Musk said Grok 3 could be released in 2024, but it surely was eventually postponed.”
In July of last 12 months, he announced through Nevertheless, in an interview in August, uncertainty was revealed by revising the discharge date to “2024, if we’re lucky.”
On this announcement, CEO Musk emphasized, “Grok 3 was pre-trained with 10 times more computing power than Grok 2.” As is thought, Grok 3 learned through xAI’s supercomputer ‘Colossus’ inbuilt Memphis, Tennessee.
Until now, the performance of generative AI models has largely trusted computing power and the quantity of coaching data. Nevertheless, recent evaluation has shown that the scaling law is blocked and that it’s difficult to attain the expected performance improvement through pre-training.
For this reason, there was speculation that Grok 3 could have missed its release deadline resulting from insufficient performance. Nevertheless, CEO Musk avoided mentioning this and easily emphasized the variety of GPUs deployed.
After all, 100,000 GPUs is currently the world’s highest level, and it’s unclear whether the boundaries of the scaling law have been revealed even with this level of computing infrastructure.
Due to this fact, if the newly introduced Grok 3 shows higher performance than the Open AI model or Google’s recently introduced ‘Gemini 2.0’, the situation may change again. On this respect, the discharge of Grok 3 is receiving considerable attention.
Moreover, CEO Musk is thought to be planning to expand the variety of Colossus GPUs by greater than 1 million, which is 10 times the present number, to coach next-generation models after the launch of Grok 3.
Reporter Park Chan cpark@aitimes.com
