Recruiting AI talent is proving to be a challenging task for many companies. The founder and CEO of Perplexity, Aravind Srinivas, shared a story that exemplifies the difficulty of hiring individuals with generative AI skills. Srinivas attempted to hire a senior researcher from Meta, but the candidate responded by saying, "Come back to me when you have 10,000 H100 GPUs." H100 GPUs are highly sought-after graphic processing units used by tech giants like Meta, OpenAI, and Google to power their AI chatbots. Acquiring such a large number of GPUs would be incredibly expensive and time-consuming.
Perplexity, which utilizes GPT-4 to power its Q&A engine, has faced challenges in finding the necessary talent to develop a large language model. Limited funds and a shortage of chips have contributed to this issue. Srinivas explained that people are reluctant to leave their current positions when they have peers to work with and access to established experimentation stacks and existing models. To attract talent, companies like Perplexity must offer exceptional incentives and immediate access to significant computational resources.
Even if smaller companies manage to obtain Nvidia's chips, they will still struggle to keep up with the rapid pace of AI development. This could make it even more difficult to secure AI talent in the future. Srinivas expressed concern that by the time companies acquire the necessary resources, the talented individuals they wish to hire will have already moved on to the next generation of models.
The demand for AI skills, particularly in areas like machine learning and data engineering, has skyrocketed since the launch of OpenAI's ChatGPT. To attract generative AI talent, companies such as Amazon, Netflix, and Meta have offered salaries as high as $900,000 per year. Even non-tech companies in sectors like education, healthcare, and law are seeking employees with AI knowledge.
While Big Tech companies may employ individuals capable of creating AI models that produce desirable outputs, Srinivas believes that this skill alone is insufficient to make AI tools truly useful. Post-training expertise, which involves addressing issues that arise when deploying AI products, is crucial. Srinivas suggests that employees from industries like crypto or e-commerce can quickly learn these skills, such as reducing factual inaccuracies in chatbots.
By focusing on post-training skills, companies like Perplexity can differentiate themselves in a market dominated by Big Tech. Srinivas believes that this approach offers a significant advantage in creating value, and his company is committed to developing expertise in this area.