technology giantswsoppokerNew direction for making money: small language model
Source: AI
Technology companies target "small models" to achieve "big gains."
After spending billions of dollars to develop large models,wsoppokerTechnology giants are turning their attention to small models, hoping to open up new paths for profit.
According to media reports on Saturday, large models had high hopes and were seen as a key driving force in promoting the development of generative AI. However, due to constraints such as high operating costs and data privacy concerns, the widespread application of this technology in the enterprise field has been hindered.
Today, technology giants such as Apple, Microsoft, Meta and Google have launched small models with fewer parameters but stronger performance, in the intention of opening up AI profit growth points.
In general, the larger the parameters, the stronger the model performance. However, large-parameter models often require the support of massive data and computing power, and their operating costs and energy consumption are high, which is undoubtedly a heavy burden for small and medium-sized enterprises.
Small models are popular for their relative cheapness, energy saving, and customization.
Eric Boyd, corporate vice president of Microsoft's Azure AI platform, said that the small model opens up more applications for customers that were previously unable to achieve due to insufficient return on investment.
The small "Gemini Nano" model released by Google this week has been built into the new Pixel phone. Meta's new 800-million-parameter Llama 3 claims to be comparable to OpenAI's GPT-4 model. French startup Mistral has also joined the camp.
At the same time, the highlight of small models is that they can ensure data privacy. These models can complete calculations locally on the device without having to upload sensitive information to the cloud, making them attractive to corporate customers who pay attention to data privacy.
In addition, small models have relatively low energy consumption, which can promote the extension of artificial intelligence from the cloud to terminal devices.
Apple released a mini model of OpenELM last month, signaling its intention to embed AI features into the best-selling iPhone line.
This article comes from "Hard AI" on the Weixin Official Accounts