在人工智能领域，一场新的技术竞争正在悄然展开，焦点集中在小模型的开发与应用。这一趋势表明，与动辄数万亿参数的大模型相比，体积更小、计算成本更低、部署更为便捷的小模型正逐渐成为AI巨头们的新战场。近期，OpenAI与Mistral AI的接连动作，无疑为这一趋势增添了新的注脚。
OpenAI的GPT-4o mini，作为GPT-3.5的替代品，不仅在ChatGPT上提供免费服务，其API价格也极为亲民，每百万输入token仅为15美分，每百万输出token为60美分，相较于之前的最便宜的GPT-3.5 Turbo模型，价格便宜了近一个数量级，且比GPT-3.5 Turbo便宜了60%以上。OpenAI CEO山姆·奥特曼对此的描述是，“通往智能的成本已经’too cheap to meter’”。这表明，随着技术的进步，人工智能服务的成本正在逐渐降低，为更多领域和用户提供可能。
与此同时，Mistral AI宣布的Mistral NeMo小模型，作为其最新成果，进一步推动了小模型在AI领域的应用。Mistral NeMo不仅继承了小模型的诸多优势，如更低的计算成本、更便捷的训练和部署，而且在满足计算资源受限、数据安全级别较高的场景方面展现出更大的潜力。这标志着Mistral AI在AI领域的小模型研发上取得了重要进展，与OpenAI等科技巨头一道，推动了小模型技术的快速发展。
这一系列事件凸显了小模型在AI领域的重要地位和潜在价值。随着技术的不断进步和成本的持续降低，小模型不仅有望在各个行业和领域得到更广泛的应用，而且可能成为未来AI发展的重要驱动力。从大模型到小模型，人工智能的未来正展现出更加多元和丰富的可能性。
英语如下：
News Title: “AI Titans in a New Battlefield: The Small Model Revolution”
Keywords: Small Models, Cost Reduction, AI Arena
News Content: In the realm of artificial intelligence, a new technological competition is quietly unfolding, with the spotlight on the development and application of small models. This trend signifies that, compared to large models with trillions of parameters, smaller models with lower costs, more convenient deployment, and smaller footprints are gradually becoming the new battleground for AI giants. Recent moves by OpenAI and Mistral AI have certainly added new layers to this trend.
OpenAI’s GPT-4o mini, a replacement for GPT-3.5, offers free services on ChatGPT and has an exceptionally affordable API price, charging only $0.15 per million input tokens and $0.60 per million output tokens. This is a magnitude lower than the cheapest GPT-3.5 Turbo model, and a whopping 60% cheaper than GPT-3.5 Turbo. OpenAI CEO Sam Altman’s description of this as “costs that are ‘too cheap to meter'” suggests that, with technological advancements, the cost of AI services is declining, making them accessible to a broader range of fields and users.
Concurrently, Mistral AI’s announcement of the Mistral NeMo small model, its latest achievement, further propels the application of small models in the AI domain. Mistral NeMo not only retains the many advantages of small models, such as lower computational costs and easier training and deployment, but also shows greater potential in scenarios where computational resources are limited and data security is paramount. This marks a significant advancement in Mistral AI’s research on small models in the AI field, alongside other tech giants like OpenAI, driving the rapid development of small model technology.
These series of events highlight the pivotal role and potential value of small models in the AI domain. With ongoing technological progress and cost reduction, small models are not only poised for wider application across various industries and fields but may also become a key driver of future AI development. From large models to small models, the future of AI is revealing a more diverse and rich array of possibilities.
【来源】https://www.jiqizhixin.com/articles/2024-07-19-3
Views: 0