TTechnology Vechnology Vision 2ision 2023 023 | When A | When Attoms meet Bitsoms meet Bits #TechVision building fbuilding foundation models aroundation models are giving thire giving third pard partietiess to-read summaries—a task that CarMax said would acceaccess thrss through APIs or bough APIs or by open soury open sourcing them,cing them, have taken its editorial team 11 years to complete. putputting theting these advse advanced models in ananced models in anyyoneone’’s hands.s hands. Other organizations are also experimenting with This is no eThis is no evvereryydaday ty technology advechnology advancement.ancement. foundation models, adapting them for tasks ranging While fWhile foundation models aroundation models are not the only are not the only areea ofa of from powering customer service bots, to generative AI rAI reeseseararch that’ch that’s grs groowing, the magnitude of theirwing, the magnitude of their product design, to automated coding. And the potpotential impact and the speed at which theential impact and the speed at which they can bey can be models are advancing fast, with companies across deplodeployyed is driving them ted is driving them to the to the top of companieop of companiess’’ industries quickly discovering new ways to use them. innoinnovvation agendas. Wation agendas. With them, busineith them, businessesses can sts can starartt As foundation models broaden and extend what we tto appro approach manoach many ty taskasks and challenges and challenges diffs differerentlyently,, can do with AI, they are letting companies transform shifshifting fting focus frocus from building their oom building their own AI twn AI to leo learningarning human-AI interaction and build an entirely new tto build o build withwith AI. generation of AI applications and services. One companOne company ty taking advaking advantantage of this is Cage of this is CarMax.arMax. It is imperative for every business to pay attention to CCarMax is using GParMax is using GPTT--33, a lar, a large language model andge language model and foundation models. To be part of this new generation prpredeceedecessor tssor to GPo GPTT--44, t, to impro improovve the care the car-buying-buying of AI applications and leverage these models to drive 174 eexperiencexperience.. Knowing that there’s a massive amount novel business solutions and offerings, companies of infof information potormation potential car buyential car buyerers mas may wy want tant too need to understand their strengths and capabilities rreead thrad through befough beforore making a pure making a purchase decision,chase decision, and track how they are advancing—starting today. CarMax used Microsoft’s Azure OpenAI Service to The bedrock of AI is shifting. Will you lose your access a pretrained GPT-3 model to quickly read and footing—or end up on higher ground? synthesize over 100,000 customer reviews for every vehicle make, model, and year that they sell. From these reviews, the model generated 5,000 easy-