Jensen Huang, CEO of NVIDIA, said on X (formerly Twitter) on September 6 that OpenAI’s GPT-6 Astra marked the arrival of “AGI.”
He also said that more than 10만 NVIDIA systems were used to train the model and that 40만 GPUs are scheduled to be brought online next.
However, this statement represents the NVIDIA CEO’s assessment and plans; it does not mean that the arrival of artificial general intelligence has been definitively confirmed.
3-Line Summary
1. Jensen Huang assessed GPT-6 as AGI
2. 10만 NVIDIA systems were used for training
3. There is still no consensus on the criteria for determining AGI
Remarks on 10만 Systems and 40만 GPUs
On X, in local time on September 6, CEO Jensen Huang wrote that GPT-6 Astra was trained using more than 10만 of NVIDIA’s Grace Blackwell NVLink72 systems. He explained that it took 4 years to progress from ChatGPT through o1 to Astra and congratulated the OpenAI team.
The figure of more than 10만 refers to the scale of NVIDIA systems used to train Astra. GPUs are core semiconductors used to train and operate AI models. As models grow larger and handle tasks involving multiple stages, the importance of the computing resources and data centers supporting them also increases.
CEO Huang went on to say that 40만 GPUs are scheduled to be brought online. However, according to an iNews24 report, he did not explain the specific purpose or deployment timing of those GPUs, or whether they represent additional capacity to be投入 into Astra. This is why it is difficult to immediately interpret the figure of 40만 as a confirmed training plan for a specific model.
“The Arrival of AGI” Is CEO Huang’s Assessment
AGI refers to artificial intelligence that can learn, reason through, and solve various problems without being limited to a specific field. CEO Huang defined Astra as AGI, but ZDNet Korea reported that opinions differ across industry and academia regarding the definition and criteria for determining AGI.
OpenAI also unveiled Astra on September 3, explaining that it had improved the model’s abilities in computer use, web browsing, software development, and scientific and professional work. According to a Joseilbo report, Astra goes beyond answering questions by strengthening its ability to operate computers and software according to user instructions and perform tasks involving multiple stages.
Improved performance and an AGI determination are not the same thing. Astra’s demonstrated high capability in certain tasks is documented, but that does not automatically lead to a conclusion about whether it can perform all intellectual activities like a human. In fact, artificial intelligence researcher Gary Marcus pointed out that there is insufficient basis for defining Astra as AGI.
The clearest fact in this statement is that NVIDIA’s large-scale computing infrastructure was used to train Astra. By contrast, the expression “the AGI era” needs to be distinguished as CEO Jensen Huang’s assessment, emphasizing the significance of that infrastructure and model.
References
Tags #JensenHuang #NVIDIA #GPT6Astra #OpenAI #AGI #ArtificialGeneralIntelligence #GPU #GraceBlackwell #NVLink72 #AIInfrastructure #DataCenters #ArtificialIntelligence