In the early 1990s, Intel was the dominant force in the CPU market, attracting tech enthusiasts and businesses with its reliable, high-performance microprocessors. However, these technologies come at a high price, making the technology financially inaccessible to many potential users, including myself.
AMD has broken this monopoly by offering affordable CPUs that rival Intel in terms of performance. By catering to an "overserved" market satisfied with "good enough" performance, AMD capitalized on the void Intel inadvertently left in the market. This shift became a textbook case for Clayton Christensen's theory of disruptive innovation. Christensen frequently references his interactions with Intel CEO Andy Grove in his numerous books and speeches. In a way, AMD's presence prompted Intel to introduce a cheaper variant of the Pentium CPU, which it sold as a Celeron variant. Celerons, while still made by Intel, have lower cache and slower clock speeds, and are designed for people (myself included) who are looking for a CPU good enough to run their computer.
Fast forward to today, and a similar shift is underway in the field of artificial intelligence (AI). Meta's motivations for open sourcing its advanced Llama2 language model may be various, but foremost among them is the democratization of large language models (LLMs), which cater to both underserved and overserved market segments.
By making state-of-the-art language models freely available, Meta is opening doors that were previously considered inaccessible. Combined with techniques like quantization that dramatically reduce CPU resource requirements, and sometimes eliminate the need for GPUs entirely, Meta is paving the way for a new, inclusive AI ecosystem.
This strategic move aims to democratize AI technology. Until recently, access to such advanced models was reserved for those who could afford them. Now, with quantization, even the CPU can host these models,