Meta为其Muse Spark AI模型推出了一项激励性定价方案，向愿意分享提示词和模型输出用于训练数据的用户提供大幅折扣1。根据该方案，贡献者可获得约95%的优惠，其中输入token价格从每百万$1.25降至$0.1，输出token价格从每百万$4.25降至$0.2，而非贡献用户仍按标准价格支付1。
这一举措反映了当前大型AI公司在获取高质量训练数据方面面临的挑战1。与此同时，AI模型的定价竞争日趋激烈，Anthropic最新发布的Fable和Mythos模型降低了缓存token的处理成本，OpenAI也在7月底大幅降低了最新模型的价格1。不过，据普林斯顿大学计算机科学教授Arvind Narayanan指出，即便消费计划提供10-20倍或更大的折扣，企业用户仍倾向于选择数据隐私保护更强的Enterprise计划1。
Meta has introduced a financial incentive program for its Muse Spark AI model, dramatically slashing prices for users willing to share their prompts and model outputs for training data 1. Contributors to the training effort receive input tokens at $0.1 per million—down from the standard $1.25 per million—and output tokens at $0.2 per million, compared to the regular rate of $4.25 per million, representing an approximately 95% discount 1.
The pricing strategy reflects broader challenges that major AI companies face in acquiring high-quality training data, even as price competition intensifies across the AI model market 1. The move echoes similar practices elsewhere in the industry: Anthropic's recently released Fable and Mythos models reduced the processing costs for cached tokens, while OpenAI significantly lowered prices for its latest models in late July 1. Meta's approach also follows internal experience with data collection efforts; the company suspended an employee computer usage tracking program in June after facing internal criticism, though it had been launched earlier in the year 1.
Data privacy considerations may influence adoption of Meta's offer. According to Princeton computer science professor Arvind Narayanan, enterprises tend to opt for plans with stronger data privacy protections, such as Enterprise tiers, even when consumer plans offer discounts of 10 to 20 times or more 1. This preference underscores the trade-offs users weigh between cost savings and data confidentiality when evaluating AI service providers.
评论
还没有评论，欢迎留下第一条。