The tech industry has long championed the idea that bigger is better when it comes to generative A.I. systems, regardless of the cost. However, companies like Microsoft are now shifting gears, recognizing that giant systems aren’t necessarily the best fit for everyone.
In a recent move, Microsoft unveiled three smaller A.I. models as part of its Phi-3 technology family, signaling a departure from the industry’s obsession with size. These compact models offer a cost-effective alternative to their larger counterparts, without sacrificing too much in terms of performance.
Even the smallest Phi-3 model performs admirably, nearly matching the capabilities of the much larger GPT-3.5 system that powered OpenAI’s ChatGPT chatbot. What’s more, this diminutive model can run on smartphones, making it accessible even without internet connectivity, and it operates efficiently on standard computer chips, avoiding the need for expensive Nvidia processors.
By requiring less processing power, these smaller models enable big tech providers to offer A.I. solutions at a lower cost, democratizing access to A.I. technology. While Microsoft claims that using these new models will be significantly cheaper than their larger counterparts like GPT-4, exact pricing details have not been disclosed.
Although these smaller systems may lack the raw power of their larger counterparts, Microsoft and other tech companies believe that many users are willing to sacrifice a bit of performance in exchange for affordability. This shift reflects a broader recognition that not all tasks require the same level of precision.
While larger, more precise A.I. systems may be necessary for certain tasks, such as medical diagnostics or financial analysis, many other applications can make do with slightly less accuracy. For instance, online advertisers aim to leverage A.I. for better ad targeting, but they need cost-effective solutions to do so regularly.
The development of smaller A.I. models represents a trade-off between power and size. Companies like Microsoft have refined their approach by focusing on improving the quality of the data fed into these models, ensuring that they learn from higher-quality text sources.
In a move towards collaboration and transparency, Microsoft has also chosen to open source its new Phi-3 models, following the footsteps of Meta and Google. This allows anyone to use and modify these models free of charge, fostering innovation and adoption across the industry.
As the tech industry continues to evolve, the emergence of smaller, more affordable A.I. models represents a significant step towards making A.I. technology accessible to a wider range of users.
Source: nytimes.com
