① Prakal Mehrotra, PayPal's senior vice president of artificial intelligence, said at the AI Summit earlier this month, “We have moved from the information age to the intelligence age.'' ②Mehrotra emphasized, “Every company needs to think about data in terms of tokens, because only then can they extract intelligence from their data.”
Cailian Press, December 26 (edited by Xiaoxiang) In the mid-20th century, the world entered the information age, the transformation of the industrial sector into information technology. This era began with the miniaturization of computers and culminated with the invention of the World Wide Web, which made information accessible to nearly everyone.
Today, with the rise of artificial intelligence, some technology leaders believe that this era is coming to an end and a new technology era is beginning.
“We have moved from the information age to the intelligence age,” Prakhar Mehrotra, PayPal's senior vice president of artificial intelligence, said at the AI Summit earlier this month.
Mehrotra explained that this “intelligence age” is characterized by industries moving away from traditional models of data storage and retrieval. Harnessing the power of artificial intelligence allows data to be generated more autonomously, with the ultimate goal of achieving automation in certain aspects of the workplace.
Companies around the world are currently racing to apply artificial intelligence in the workplace to increase productivity and output, with mixed results. A study conducted by MIT in August showed that 95% of enterprise-level AI office initiatives fail to achieve rapid revenue growth.
“This will be a journey…one will have to go through the 'crawl, walk, run' stages,” Mehrotra pointed out. “This adage was true 10 years ago and it's still true today.”
Marc Hamilton, NVIDIA's vice president of solution architecture and engineering, who spoke with Mehrotra, said the key to how companies build AI systems in the future is to invest in “AI factories,” whether deployed on-premises or in the cloud. This is because the data needed to run a company will no longer be primarily obtained by humans or computers, but will be generated by AI.
“If you're asked, 'PowerPoint slide with specific content,' or 'I'm working on this coding feature, can you generate the code for me?'” — this doesn't pull the information from the database. Calling the model to generate data. ” Hamilton explained.
Tokens determine success
Mehrotra noted that for companies to effectively build the computing power needed to generate such data, they need to focus on a new type of basic unit: the token. The basic unit for text-based AI to understand and process language, tokens are both pieces of training data and output content produced by a model after receiving instructions.
“Every company needs to think about their data in terms of tokens, because only then can they extract intelligence from their data,” Mehrotra emphasized.
As a metric for measuring input and output, the amount of tokens generated has become an important metric for technology companies. In May of this year, NVIDIA announced that its chip client Microsoft generated over 100 trillion tokens in the first quarter. This is a 5x increase compared to the previous year. These production figures help AI companies market themselves to investors and boost valuations, but the data suggests that the correlation between tokens and demand or profitability is weaker than what tech companies suggest.
Mehrotra and Hamilton both believe that many companies recognize the value of tokens in enhancing their artificial intelligence capabilities, but are still considering how best to integrate tokens with their specific needs. All companies operate some form of AI factory, both receiving and producing tokens of value.
“I think of it as a type of 'strength training,'” Mehrotra says. “If every employee started thinking in terms of tokens and generation processes, then yes, this would be a completely different company.”
