广告
加载中

识假店 生成后厨预警!淘宝闪购研发食品安全AI大模型

亿邦动力 2026-02-28 16:50
亿邦动力 2026/02/28 16:50

邦小白快读

EN
全文速览

淘宝闪购推出免费AI大模型“白泽”,专为食品安全和风控设计,提供实用监管工具。

1. 模型能24小时自动分析后厨视频流,识别关键细节如墙面清洁度、垃圾桶外溢、生熟食材分离和厨师工作帽佩戴,确保食品安全。

2. 防范虚假店铺,通过多图比对和思维推理技术验证商户环境与工商信息,自动标记矛盾点,遏制“一店多开”和证照不符行为。

3. 商家可上传“一镜到底”视频申请“可堂食”打标,模型跨帧分析视频真伪,防止AI造假,保障消费者真实体验。

4. 技术高效且低成本,部署成本仅同类方案1/20,已接入100多个场景,调用量突破10亿,Token消耗超万亿,适合大规模落地。

该模型帮助品牌商提升消费者信任,优化品牌形象。

1. 通过后厨监控功能,品牌可展示透明厨房操作,强化食品安全承诺,迎合消费趋势中对卫生和透明的需求。

2. 虚假店铺识别减少品牌声誉风险,确保线上线下服务一致性,提升用户行为观察的可靠性。

3. 开源免费模型允许品牌总部部署管理分店合规,降低成本,为产品研发提供数字工具启示。

4. 代表企业淘宝闪购的案例证明,低成本AI应用能增强品牌渠道建设,吸引注重安全的消费者。

模型带来增长机会和风险应对方案,支持卖家合规经营。

1. 免费开源技术提供扶持政策,卖家可低成本部署,上传视频打标获取“可堂食”认证,提升店铺吸引力。

2. 识别后厨违规和虚假证件,降低风险提示,避免平台处罚,确保正面影响如消费者信任增强。

3. 适应消费需求变化,AI全时监控替代人工抽查,提高效率,抓住食品安全需求增长的市场机会。

4. 最新商业模式启示:可借鉴淘宝闪购做法,合作方式如平台集成模型,Token消耗数据(超万亿)验证可靠性。

模型启示工厂推进数字化生产,探索商业机会。

1. AI技术可应用于产品生产环境监控,如清洁度和材料分离,确保食品安全标准设计需求。

2. 低成本部署(仅同类1/20)和开源特性,工厂可大规模定制,提升电商启发的效率。

3. 商业机会包括模型训练适应工厂场景,性能数据(调用量10亿)展示稳健应用潜力。

4. 案例中代表企业阿里云的技术支持提供推进数字化启示,Token消耗(万亿级别)证明可行性。

模型代表行业发展趋势,解决客户痛点。

1. 新技术如多模态大语言模型,针对后厨预警和虚假店铺检测提供解决方案,满足客户对智能监控的需求。

2. 高性能优势,在强噪声和高异质性数据中保持稳健,技术数据如Token突破万亿验证可靠性。

3. 开源Apache 2.0协议允许服务商自由修改和部署模型,整合到解决方案中,推动行业创新。

4. 代表案例淘宝闪购的100多个生产场景接入,展示应用广度,调用量10亿级别证实可扩展性。

模型提供平台治理最佳做法,降低风险。

1. 淘宝闪购最新做法:AI全时监控替代人工抽查,处理商家入驻审核、日常巡检和风险预警,提升运营管理效率。

2. 解决如虚假店铺问题,通过多图比对技术规避一店多开风险,增强平台安全性。

3. 免费开放技术,平台商可借鉴用于招商和运营管理,调用量10亿数据支持大规模应用。

4. 开源模型允许平台部署低风险方案(成本仅1/20),Token消耗超万亿确保决策稳健。

模型揭示产业新动向,开源模式提供政策启示。

1. AI在食品安全治理的应用是新动向,技术性能超越基准,在真实门店数据中保持稳健感知。

2. 开源Apache 2.0协议鼓励商业创新,免费使用无需许可,提供政策法规建议如专利许可框架。

3. 商业模式分析:低成本高效果(成本1/20),调用量10亿和Token万亿数据展示推广潜力,研究新问题如多模态决策效率。

4. 代表研究者夏威的项目数据提供案例参考,Token消耗级别支持学术验证。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

Taobao's flash sale platform has launched a free AI model named "Baize," specifically designed for food safety and risk control, offering practical regulatory tools.

1. The model can automatically analyze back-of-house video streams 24/7, identifying key details like wall cleanliness, overflowing trash bins, separation of raw and cooked ingredients, and chefs' hat usage to ensure food safety.

2. It prevents fake stores by comparing multiple images and using reasoning techniques to verify merchant environments against business registration data, automatically flagging inconsistencies to curb practices like "one location, multiple stores" and mismatched licenses.

3. Merchants can upload continuous, unedited videos to apply for a "Dine-In Available" label; the model analyzes video authenticity across frames to prevent AI-generated fraud, safeguarding genuine consumer experiences.

4. The technology is efficient and low-cost, with deployment expenses only 1/20th of comparable solutions. It's already integrated into over 100 scenarios, handling over 1 billion calls and consuming trillions of tokens, demonstrating suitability for large-scale implementation.

This model helps brands enhance consumer trust and optimize brand image.

1. Through back-of-house monitoring, brands can showcase transparent kitchen operations, reinforcing food safety commitments and aligning with consumer demand for hygiene and transparency.

2. Fake store identification reduces reputational risks, ensures consistency between online and offline services, and improves the reliability of user behavior observations.

3. The open-source, free model allows brand headquarters to deploy and manage branch compliance, lowering costs and providing inspiration for digital tool development in product R&D.

4. Taobao Flash Sale's case demonstrates that low-cost AI applications can strengthen brand channel development and attract safety-conscious consumers.

The model offers growth opportunities and risk mitigation solutions, supporting compliant operations for sellers.

1. Free, open-source technology provides support policies, enabling sellers to deploy it at low cost, upload videos for "Dine-In Available" certification, and enhance store appeal.

2. It identifies back-of-house violations and fake credentials, reducing risk alerts, avoiding platform penalties, and ensuring positive impacts like increased consumer trust.

3. Adapting to changing consumer demands, AI's round-the-clock monitoring replaces manual checks, improving efficiency and capitalizing on the growing market for food safety.

4. Latest business model insights: Sellers can learn from Taobao Flash Sale's approach, such as platform integration of the model, with token consumption data (exceeding trillions) validating reliability.

The model inspires factories to advance digital production and explore commercial opportunities.

1. AI technology can be applied to monitor production environments, such as cleanliness and material separation, ensuring food safety standards in design requirements.

2. Low-cost deployment (only 1/20th of comparable solutions) and open-source features allow factories to customize at scale, improving efficiency inspired by e-commerce.

3. Commercial opportunities include adapting model training for factory scenarios, with performance data (1 billion calls) demonstrating robust application potential.

4. The case study of Alibaba Cloud's technical support offers digital advancement insights, with token consumption (trillions) proving feasibility.

The model represents industry trends and addresses client pain points.

1. New technologies like multimodal large language models provide solutions for back-of-house alerts and fake store detection, meeting client demand for intelligent monitoring.

2. High-performance advantages include robustness in noisy, heterogeneous data environments, with technical data like trillions of tokens validating reliability.

3. The open-source Apache 2.0 license allows service providers to freely modify and deploy the model, integrating it into solutions to drive industry innovation.

4. Taobao Flash Sale's integration into over 100 production scenarios showcases application breadth, with 1 billion calls confirming scalability.

The model offers best practices for platform governance and risk reduction.

1. Taobao Flash Sale's latest approach: AI's 24/7 monitoring replaces manual checks, handling merchant onboarding reviews, daily inspections, and risk alerts to improve operational efficiency.

2. It addresses issues like fake stores through multi-image comparison technology, mitigating risks such as "one location, multiple stores" and enhancing platform security.

3. The free, open technology can be adopted by platforms for merchant recruitment and management, with 1 billion call data supporting large-scale application.

4. The open-source model allows platforms to deploy low-risk solutions (costing only 1/20th of alternatives), with trillions of tokens consumed ensuring robust decision-making.

The model reveals new industry trends, with its open-source approach offering policy insights.

1. AI's application in food safety governance is an emerging trend, with technical performance surpassing benchmarks and maintaining robust perception in real-store data.

2. The Apache 2.0 open-source license encourages commercial innovation, allowing free use without permits and providing policy suggestions like patent licensing frameworks.

3. Business model analysis: Low cost, high effectiveness (1/20th the cost), with 1 billion calls and trillions of tokens demonstrating scalability, raising new research questions like multimodal decision efficiency.

4. Representative researcher Xia Wei's project data offers case references, with token consumption levels supporting academic validation.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

【亿邦原创】2月28日消息,亿邦动力获悉,日前淘宝闪购发布专为餐饮服务与零售门店打造的风控治理垂直领域开源大模型——“白泽”(Ostrakon-VL),并宣布相关技术能力向全行业免费开放。

据悉,借助“白泽”大模型,外卖平台、餐饮及零售企业在图像识别、后厨预警等方面,都能通过人工智能技术快速加强对不合规场景的识别和治理能力,提高数字化治理效率。

作为首个面向餐饮服务与零售门店场景的多模态大语言模型,“白泽”已经深入参与到了淘宝闪购的日常平台治理中。比如,在商家入驻审核、日常巡检及风险预警等核心场景中发挥关键作用。

在明厨亮灶智能巡检方面,“白泽”可以24小时分析后厨直播视频流,自动识别墙面、台面清洁度、垃圾桶是否外溢、生熟食材是否分离、厨师是否佩戴工作帽等食品安全管理细节。白泽识别出违规行为后,系统可以即时生成预警并推送至风控治理团队,实现从“人工抽查”到“全时监控”的转变。

针对虚假店铺问题,“白泽”通过多图比对与思维推理技术,将商户环境与工商信息进行比对,模型可自动标记矛盾点,有效遏制“一店多开”和证照不符等违规行为。

目前,淘宝闪购已支持商家通过上传“一镜到底”的视频,来为其进行“可堂食”打标。“白泽”也能根据视频内容及店铺相关信息,综合判断店铺是否具备真实堂食环境,通过跨帧一致性分析识别视频真伪,防止用静态图片或AI生成视频蒙混过关,确保消费者获得真实的服务体验。

据“白泽”项目负责人夏威介绍,该模型基于阿里云通义千问Qwen3-VL-8B基座进行深度训练调优,针对餐饮、零售门店场景的特殊需求进行了垂直领域适配,在技术性能方面的主要优势是极高的参数效率。根据ShopBench(服务于餐饮和零售门店的公开基准)评测,“白泽”性能评分显著超越同规模基座模型,在强噪声、高异质性的真实门店数据中仍能保持稳健的感知与决策能力。

目前,“白泽”已经接入淘宝闪购超过100个生产场景,调用量突破10亿级别,Token消耗量已突破万亿级别。更重要的是,“白泽”的部署成本仅为同类方案的1/20,有助于网络餐饮行业的各参与方以极低成本大规模落地应用。

为推动行业整体食品安全治理水平提升,淘宝闪购宣布“白泽”相关技术能力向行业免费开放。模型权重与技术报告已通过主流开源平台发布,采用Apache 2.0开源协议。

据悉,该协议允许任何机构在商业或非商业场景下免费使用,无需支付许可费用,修改后的代码可闭源分发,并明确规定贡献者授予全球性、免版税的专利许可。使用方可根据需求自由部署和修改模型,实现“一次开源,多方受益”。

通过开源开放,连锁品牌总部可利用模型管理分店合规情况;各大外卖平台也可借鉴淘宝闪购技术方案提升自身治理能力。淘宝闪购技术相关负责人表示,平台长期以来持续加大AI研发投入,让消费者放心消费,让合规商家获得更有利的经营环境。

亿邦持续追踪报道该情报,如想了解更多与本文相关信息,请扫码关注作者微信。

文章来源:亿邦动力

广告
微信
朋友圈

这么好看,分享一下?

朋友圈 分享

APP内打开

+1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0