Generative AI reaches a crossroads. Where is the next wave?

Generative AI is becoming more and more popular, especially in the business world. Not long ago, Walmart announced the launch of generative AI applications for use by 50,000 non-store employees. The app combines Walmart data with third-party large language models (LLM) to help employees perform a variety of tasks, such as becoming creative partners and extracting summaries from large documents.

Due to the popularity of generative AI, the demand for GPUs has increased, and training deep learning models requires powerful GPUs. According to the Wall Street Journal, training AI models can cost billions of dollars because of the massive amounts of data that need to be processed and analyzed.

New trends have brought considerable business opportunities to NVIDIA, and NVIDIA GPU has become a hot money-making machine. In order to obtain Nvidia chips, startups and investors take extraordinary measures. The "New York Times" column stated: "Compared with money, engineering talent, hype and even profits, companies seem to need GPUs more this year."

In this possible technological change, Nvidia stands at the top of the mountain. At this time, Google reached a cooperation with NVIDIA to provide technical support based on NVIDIA GPUs to Google Cloud customers. Does the current surge in demand mean that generative AI has reached its peak, or is it the beginning of the next wave? This is a question that everyone is thinking about.

At the recent earnings call, Nvidia CEO Jensen Huang pointed out that increased demand marks the beginning of accelerated computing, and it is just the dawn. Huang Renxun suggested that enterprises should reallocate investments and not just focus on general computing, but should pay more attention to generative AI and accelerated computing.

General purpose computing refers to CPU-based computing, but NVIDIA believes that CPU has become a backward infrastructure, and developers should optimize for GPU because GPU is more efficient than traditional CPU. GPU can process multiple calculations in parallel at the same time, making it particularly suitable for deep learning. GPUs also have unique advantages when dealing with certain mathematical problems, such as linear algebra and matrix operation tasks.

Unfortunately, many software are only optimized for CPU and cannot benefit from GPU parallel computing. In the future, many CPU tasks will be performed by GPUs, which is an opportunity for Nvidia, because generative AI will generate massive amounts of content and requires cloud computing support.

Human beings and businesses are lazy. Now that the software has been optimized for the CPU, they are unwilling to invest resources and time in the GPU.

When machine learning first emerged, data scientists were too ambitious and wanted to apply it to everything, even if simpler tools already existed in some fields. To be honest, machine learning can solve only a very small number of business problems. In short, accelerated computing and GPU are not suitable for all software.

To welcome the next wave, generative AI needs to break through

Looking at the current situation, Nvidia’s performance data is indeed eye-catching, but Gartner warns that generative AI is at a The peak of anticipated inflation. Some assert that generative AI hype has devolved into unfounded excitement and exaggerated expectations.

The generative AI craze may soon hit a bottleneck. SK Ventures venture capitalists believe: "We have now entered the long-tail stage of the first wave of large language model AI. The wave started in 2007, when Google released a paper called "Attention is All You Need". In the next 1-2 years, everyone will hit a bottleneck." What are the bottlenecks? Such as the tendency to hallucinate, insufficient training data in a narrow domain, the aging of the training corpus from many years ago, and countless other factors. In short, we are now most likely entering the tail end of the first wave of AI.

Does this mean that generative AI is about to die? No, it just means that generative AI requires major technological breakthroughs, so that productivity can be greatly improved and better automation can be fostered. In the next wave of generative AI, new models, more openness, and ubiquitous cheap GPUs may be the key.

The long run should be bright for generative AI, as labor is in short supply and humans need better automation technology. Looking back at history, AI and automation seem to be two independent technology categories, but generative AI has changed this view. Workflow co-founder Mike Knoop said: "AI and automation are collapsing into the same thing." McKinsey said in the report: "Generative AI will breed the next great improvement in productivity." Goldman Sachs believes that generative AI can increase global GDP Increased by 7%. (Knife)

The above is the detailed content of Generative AI reaches a crossroads. Where is the next wave?. For more information, please follow other related articles on the PHP Chinese website!