In the fast-paced race to develop generative A.I. systems, the tech industry has prioritized bigger systems regardless of cost.
However, tech companies are now shifting towards smaller A.I. technologies that are cheaper, though less powerful. This trade-off may be appealing to many customers.
Microsoft recently unveiled three smaller A.I. models as part of their Phi-3 technology family. The smallest model performed almost as well as GPT-3.5, a larger system that powered OpenAI’s ChatGPT chatbot.
The smallest Phi-3 model can run on smartphones without internet connectivity and on regular computer chips instead of pricier Nvidia processors.
Due to lower processing requirements, big tech companies can offer smaller models at a lower cost to customers. This could enable broader application of A.I. in areas where larger models were previously too expensive.
While smaller systems may be less accurate or sound less natural, the affordability factor is driving the adoption of these models by customers.
Microsoft’s Eric Boyd noted that smaller models are more cost-effective to deploy, making A.I. accessible to a wider range of users.
Different industries have varying needs for A.I., with some tasks requiring precision while others can tolerate slight inaccuracies. Lower costs make A.I. more accessible for regular use.
Chatbots rely on large language models (L.L.M.s) that analyze vast amounts of text to generate responses. Retrieving information from these models is computationally intensive and costly.
Tech giants like Meta and Google are also developing smaller A.I. models to reduce costs. Some companies have made these models open source to encourage industry-wide adoption and improvement.
Researchers are finding ways to make smaller models perform comparably to larger ones by optimizing data input and refining text quality.
Microsoft has introduced three different small models under the Phi-3 series, each varying in size and power. Making A.I. systems compact enough for smartphones and personal computers can significantly reduce costs.
