The CEO of an AI startup, Aravind Srinivas, recently shared his struggles in hiring AI talent due to a lack of compute power. In a podcast interview, Srinivas revealed that he was unable to poach a Meta researcher because his company, Perplexity, did not have enough GPUs to meet the candidate’s requirements.
Srinivas highlighted the challenges of recruiting workers with generative AI skills, emphasizing the need for “amazing incentives” to attract top talent in the industry. He explained that the high demand for GPUs, particularly Nvidia’s H100 GPUs, used by tech giants like Meta and Google, has made it difficult for smaller firms like Perplexity to compete.
Despite the chip shortage and limited funds, Perplexity powers its Q&A engine using GPT-4 but struggles to find the talent needed to create a large language model. Srinivas emphasized the importance of offering immediate availability of compute power and attractive incentives to entice AI talent to join smaller companies.
The rapid development of AI technology poses a challenge for companies like Perplexity, as major tech companies continue to advance at a fast pace. Srinivas expressed concerns that by the time smaller firms acquire the necessary resources, larger companies will have already moved on to the next generation of AI models.
In response to the increasing demand for AI skills, companies like Amazon, Netflix, and Meta have offered lucrative salaries to attract generative AI talent. Srinivas believes that post-training expertise, such as addressing issues like factual inaccuracies in AI models, is crucial for making AI tools useful in real-world applications.
By focusing on post-training skills and leveraging their expertise in areas like reducing chatbot inaccuracies, Srinivas believes that companies like Perplexity can differentiate themselves in a competitive industry dominated by Big Tech. Despite the challenges of hiring AI talent, Srinivas remains optimistic about the value that AI companies can create by emphasizing post-training skills and addressing the long tail of issues in AI development.