When a tool as powerful as generative artificial intelligence becomes available at large scale, five-year business plans go right out the window.

“You can practically hear the shrieks from corner offices around the world as people try to figure out how to do this,” Mat Honan (pictured, right), editor in chief at MIT Technology Review, said in his opening remarks Monday during the Future Compute conference on the MIT campus in Cambridge, Massachusetts. “It’s a tough time to be forecasting where and how to transform your business. I genuinely can’t remember a time when things have been more uncertain than they are today.”

That uncertainty is being driven by rapidly growing interest in OpenAI LP’s ChatGPT and its generative AI technology. The engine’s ability to process simple commands for loan origination, passing medical licensing exams or generating computer code has led companies across nearly every industry to ponder what this will mean going forward. The future of compute has suddenly taken a detour into a dense forest with no map.

Role of Nvidia GPUs

OpenAI has captured the world’s attention with its robust generative AI model, and one of the companies that has played a central role in making this a reality is Nvidia Corp.

To build ChatGPT, OpenAI needed access to supercomputer technology for training its models. The company formed a partnership with Microsoft Corp. in 2019 to leverage Azure cloud resources, which included thousands of Nvidia Corp. AI-optimized graphics processing units linked through a low-latency network based on Nvidia Quantum InfiniBand.

“We at Nvidia are massive fans of OpenAI,” Manuvir Das (left), vice president of enterprise computing at Nvidia, said in a presentation during the conference. “We helped them build their whole technology.”

Das is intimately familiar with the engine behind OpenAI, having designed the programming model for Azure between 2002 and 2012, before moving on to executive positions at Dell EMC and Nvidia. Das believes that the massive data sets and common access to information contained in ChatGPT will lead to fine-tuning and custom tailoring of models. This process for crafting effective instructions to obtain desired outcomes from AI is known as prompt engineering.

“Every company is going to need its own model,” Das said on Monday. “Your prompt tunes it every time you use it. As you guide it, you get a better and better answer. You are going to see a cottage industry of people who get that training.”

Building guardrails

Although fine-tuning may create more useful enterprise models, it still doesn’t resolve concerns around security and potential misuse of the generative AI tool. To help address this situation, Nvidia released NeMo Guardrails into open source last month.

Guardrails is a framework for developers to keep AI chatbots safe and accurate. It applies simple rules set by the developer to ensure that a bot’s responses are accurate and appropriate.

In a separate interview with SiliconANGLE, Das described the importance of this solution and why it was necessary in a new world where massive amounts of information are suddenly available to all. This may work fine in the preparation of a college essay, but it can create real problems in an enterprise setting.

“One of the first things that comes up in conversation is this aspect of safety and security,” Das said. “Should it give the answer to that employee? You have to think about how this can be used and not misused.”

Amazon CTO Werner Vogels

A need for guardrails around generative AI is becoming a recurring theme in comments from key leaders within the tech world. It was a major topic of discussion during the RSA Conference in San Francisco last week and was on the mind of Werner Vogels, vice president and chief technology officer at Amazon.com Inc., who spoke at Future Compute.

Vogels described how he asked ChatGPT to write a document that would narrate the influence of cybercrime on the cloud industry. It came back with an analysis that outlined how the cloud industry was entering a period of decline. “The article looked amazing, well-written, well-researched, and not the truth,” Vogels said. “There are lots of tasks where humans still need to make the decisions, not the machines.”

Vogels, who makes a set of predictions for the tech world at the end of each year, offered a bonus forecast surrounding generative AI on Monday for conference attendees. He echoed Das’ point that individual companies will fine-tune custom models and predicted that the field will see rapid innovation in user interfaces for AI.

Custom chip investment

Vogels also indicated there will be a continuation of custom chip development, an initiative his company and others, notably Apple Inc., have actively pursued. Amazon Web Services Inc. has deployed its Nitro system to provide hardware and software components on Arm chips for EC2 instances. AWS Graviton processors have been custom built for cloud workloads, and Apple has developed its own M2 system-on-a-chip line. “Custom silicon is the way to go,” Vogels said.

The rise of generative AI is likely to place more pressure on chip manufacturers to meet the technology’s power-hungry processing demands. This comes at a time when the U.S. semiconductor industry received significant financial backing from its government with the passage of the $280 billion Chips and Science Act in 2022, including $52 billion for processor research and manufacturing.

More than 200 companies have expressed interest in applying for semiconductor funding, according to the Commerce Department. However, Chris Miller, associate professor of international history at Tufts University and author of “Chip War,” has questioned whether the opening of new fabrication plants as a result of the government investment will have a lasting impact.

“Will the companies that have opened those facilities as a result of the CHIPS Act keep investing in the long run?” Miller asked. “That remains to be seen.”

Rising questions about AI and its potential impact on jobs and cybercrime are complicating the picture for the tech industry as a whole. Tesla Inc. and Twitter Inc. Chief Executive Elon Musk recently called for regulatory oversight of AI and the Biden administration has begun soliciting public comment on potential accountability measures for the technology.

“The cloud computing industry and chip industry are not at all ready when political systems come knocking on their doors and ask about what they are going to do about regulating AI,” Miller said.

Ready or not, the tech world has been forced to focus on a host of issues surrounding the AI tsunami. There may be a major amount of uncertainty, as described by Honan, but there is also a sense of responsibility in some quarters to own a solution.

“We as technologists are the stewards of this future,” Vogels told the MIT gathering. “Transparency is paramount.”

Photos: Mark Albertson/SiliconANGLE