档案库 · AI 与模型 · 产品决策 · 2016
Scale在YC S16 Demo Day推出,押注开发者会为人工劳动调用API
由Quora和Snapchat校友Alexandr Wang和Lucy Guo创立的Scale,在YC S16上作为一款API推出,提供经过筛选的按需人工,处理AI尚无法完成的工作。
Scale
做的是什么生意
Scale offers a REST API that developers drop into an app to request on-demand human work — categorization, content moderation, comparison, transcription and phone calls — done by screened, hourly-paid contractors with quality checks, competing with Mechanical Turk.
启动资金:A small seed round from Arena VC plus a place in Y Combinator's Summer 2016 batch; coverage disclosed no amount.
起因
Co-founders Alexandr Wang and Lucy Guo knew the problem from Quora and Snapchat, where manual content moderation of user-flagged images and posts was constant work, and from their first startup, a mobile app for booking doctor's appointments, which stalled because they spent all day calling doctors instead of building. That frustration became Scale: an API for the human-powered functions behind software.
经过
Scale launched officially on 2016-07-25, three weeks old, freemium with five free requests and usage tiers after that, already processing 20,000+ API calls with customers Houzz, HigherMe, Hush and RealTalk plus seven pilots. The founders framed it as the reality behind the chatbot boom: assistants like Facebook's M and Magic looked AI-powered but leaned on human helpers because machine learning was not good enough yet. At YC S16 Demo Day 1 on 2016-08-22, TechCrunch reported Scale growing 40% week over week with over 50% gross margins, with Houzz using it to screen listings and Teespring using it for moderation and categorization.
还没有结局,它还在跑。
背景
Scale由Alexandr Wang和Lucy Guo创立,于2016年7月推出,作为按需人工劳动的REST API:开发者添加一行代码,Scale经过筛选、按小时计酬的合同工就能处理分类、内容审核、比较、转录或电话呼叫,并在结果返回前进行质量检查。创始人是在Quora和Snapchat进行了人工审核,以及在他们的第一个创业项目(一个医生预约应用)因打电话的时间成本过高而失败后,创建了这家公司。
他们的主张是,Mechanical Turk的开放式众包质量低下,而业务流程外包无法通过代码调用。Scale筛选工人,按小时支付约为Turk两到三倍的工资,以免他们匆忙工作,并通过标准开发者API提供质量检查的结果。联合创始人Lucy Guo认为,所谓的AI助手就是证明:Facebook的M和Magic之所以有效,是因为背后有人类,因为机器学习当时还不够好。
Scale推出仅三周,就处理了超过20,000次API调用,客户包括Houzz,并进入YC S16 Demo Day 1,于2016-08-22报告每周增长40%,毛利率超过50%,Houzz筛选房屋列表,Teespring审核用户内容。该公司的赌注是,开发者会将人工视为可调用的基础设施,直到机器能胜任工作。
这件事要成立,得有什么
- 创始人亲身经历过这种痛苦:Quora和Snapchat进行人工内容审核,而他们的第一个创业公司因给医生打电话而非开发产品而夭折。
- Mechanical Turk的质量差距是突破口:开放的众包允许任何人参与,导致输出质量低下,而Scale筛选工人,按小时支付2-3倍于Turk的工资,并增加质量检查。
- 聊天机器人的繁荣提供了时机:像Facebook的M和Magic这样的“AI”助手实际上依靠人类运行,证明了通过API提供后台人工劳动的需求。
- 将工人定位为API使Scale成为开发者基础设施:只需一行代码而非供应商磋商,这也是为什么YC S16上每周增长40%,毛利率超过50%。
可借鉴之处
兜售炒作背后的人工:2016年每个“AI”产品都秘密依靠人工运行,因此Scale将经过筛选的工人打包成带有质量检查的API——这是AI日后大规模发展所需的基础层。
后续进展
截至2016-08-22,Scale已于2016-07-25公开发布,最初几周处理了超过20,000次API调用,客户包括Houzz、HigherMe、Hush和RealTalk,并在YC S16 Demo Day 1上展示了其每周增长40%和毛利率超过50%的业绩,Houzz和Teespring是其重要用户。
资料来源
- All 44 startups that launched at Y Combinator S16 Demo Day 1
- Scale lets companies outsource their human-powered tasks with one line of code
发现哪里写错了?告诉我们。
轮到你了
你刚读完一家。说说你在做什么,看看谁在赌同一件事。
免费账号 · 3 次免费提问 · 不用绑卡