This text initially appeared on Enterprise Insider.
Recruiting AI expertise is usually a powerful feat for some firms.
Aravind Srinivas, the founder and CEO of Perplexity, an AI-powered question-and-answer engine, described his interplay with a job candidate that exhibits how onerous it may be to rent folks with generative AI expertise.
“I attempted to rent a really senior researcher from Meta, and you understand what they stated? ‘Come again to me when you could have 10,000 H100 GPUs,'” Srinivas stated on a latest episode of the enterprise recommendation podcast “Make investments Just like the Finest.”
H100 GPUs consult with Nvidia’s extremely coveted graphic-processing items that tech giants like Meta, OpenAI, and Google use of their information facilities to energy and prepare their AI chatbots.
“That will price billions and take 5 to 10 years to get from Nvidia,” Srinivas stated.
Restricted funds, mixed with a chip scarcity, means Perplexity, which powers its Q&A engine utilizing GPT-4, has discovered it powerful to search out the expertise required to create a big language mannequin, Srinivas stated.
Srinivas stated it is troublesome to get workers to go away an organization the place they “have an important experimentation stack and present fashions to bootstrap from.”
“You must supply such superb incentives and instant availability of computing. And we’re not speaking of small compute clusters right here,” he stated.
The CEO added that even when smaller companies like Perplexity are in a position to get Nvidia’s chips, they will proceed to fall behind as a result of AI is creating so shortly.
Srinivas stated AI expertise at main tech firms “can have already made the next-generation mannequin.”
“They’re like, ‘Look, the world has modified, I am already within the subsequent era,'” he added. “‘I will come when the following model of the mannequin is completed coaching. This time, you come again to me when you could have 20,000 H100s.'”
Srinivas and Meta did not instantly reply to a request for remark from Enterprise Insider earlier than publication.
There’s been a fast uptick in curiosity in AI expertise like machine studying and information engineering since OpenAI launched ChatGPT in November 2022. Firms like Amazon, Netflix, and Meta have provided salaries as excessive as $900,000 a 12 months to draw generative AI expertise, and non-tech firms throughout the schooling, healthcare, and authorized sectors have been trying to fill roles with employees who know how one can use AI.
Srinivas believes that employees want expertise past the flexibility to create AI fashions that generate fascinating outputs.
“You must post-train them and deal with the lengthy tail of points you get on serving a product,” the CEO stated.
Put up-training experience, like realizing how one can scale back a chatbot’s factual inaccuracies, is a crucial ability that workers from a variety of digital industries can study shortly, Srinivas stated.
Leaning into that ability set, he stated, will assist AI firms like Perplexity stand out in a sector dominated by Massive Tech.
“You may have great benefit to create a variety of worth,” he stated about post-training expertise. “And we’re centered on that.”
