Artificial Intelligence (AI) is definitely having a moment– but did you know that KIOXIA has been leveraging this technology’s game-changing capabilities for a long time now?
It’s true. You might be surprised to learn that KIOXIA has been publishing research on AI for years, and has been using AI in our manufacturing processes since 2017.
Our earliest use of AI was centered on machine vision technologies, which was used to look for defects in the wafers that our flash memory chips are made with as they progressed through the manufacturing process. More recent research from KIOXIA has been focused on the area of Approximate Nearest Neighbor Search (ANNS), which is of particular importance to vector database solutions used in Retrieval Augmented Generation (RAG). RAG allows solutions to leverage Large Language Models (LLMs) for generative AI, taking advantage of domain or company specific data that may be proprietary and confidential.
Timing is Everything: Why AI is Really Taking Off Right Now
Artificial intelligence has been around for decades at this point, yet recently it has exploded in terms of its growth. There are, of course, many hardware advancements and algorithmic breakthroughs that have enabled this growth, however I believe that the current era of AI is being defined by LLMs and generative AI. By way of definition, an LLM is a type of artificial intelligence that can understand and generate human language. It works by processing vast amounts of text data (typically scanning the public data available on the internet), learning patterns and relationships between words and sentences. This allows the model to predict the next word in a sequence, generate text, translate languages, and answer questions.
Specifically, the progress made in LLMs is making AI applications available with natural language interfaces – which, in my opinion, more than any other single factor is responsible for the phenomenal proliferation of AI based solutions we are experiencing today.
What has enabled this?
Well, from a computational standpoint modern GPUs are able to perform at levels that were in the realm of supercomputers just a few years ago (as an aside most of today’s supercomputers rely heavily on GPUs). Additionally, new algorithms such as transformer architecture developed by Google have significantly advanced the art. That said, I’ll advance a somewhat unintuitive opinion: the availability of large amounts of data to train on is what has been the tipping point for the current AI growth trend. The sheer volume of data available on the internet was key to the training of the LLMs that are powering the current growth of the AI market. LLMs are fundamentally what is enabling AI solutions to be utilized by people who are not trained as technical experts by allowing a natural language interaction with AI systems. This massively broadens the reach and applicability of AI-based solutions. The current generation of LLMs are quite good, and in the near future - as they improve further - there will be very few aspects of our lives that are not directly affected by AI systems.
Flash Memory: Making it All Possible
In case it isn’t obvious by now, data is the lifeblood of AI systems. Massive amounts of data are required to train models, and as we will cover in an upcoming blog, preparing domain specific data or private proprietary data for use by LLMs expands the size of the data. To efficiently utilize precious GPU resources – and to provide quick responses to queries - requires high performance storage devices.
Flash technology, which was invented by KIOXIA, is the key to providing the bandwidth and random-access lookups required by AI solutions. Modern AI systems would simply not be possible without flash memory. AI is a rapidly changing field, and the demands placed on storage are always increasing. KIOXIA is actively working on both internal research and ecosystem collaborations to address the future storage needs of AI.
Stay tuned – this is only the first in a series of AI-focused blogs that will cover everything from the storage challenges facing AI to KIOXIA’s exciting, novel ANNS research and much more.
Notes:
All other company names, product names and service names may be trademarks of their respective companies.