https://www.reddit.com/r/technology/comments/p3od64/zillow_other_tech_firms_are_in_an_arms_race_to/ 这里面的讨论也是很开眼的 The consequences are foreseen. People with middle class wages or less will be forced to rent indefinitely. Many will have savings for a down payment but will not be able to close due to being out bid by straight cash. The ones that do manage to get in will be stuck waiving contingencies and likely get wrecked from underlying issues on the overvalued home. It just goes on and on. A buddy just bought, even had a home inspector, but still got walloped by foundation issues. Inspector saw some issues, but didn''t seem to think they were major, so my friend desperate for a house pulled the trigger. No clue how he''ll afford the foundation issues without a second mortgage, if he can qualify. Your scenario is already playing out for a lot of people. 我记得之前huaren上就有人说过 硬伤房拿出来卖的最佳时机！ 然后还有这个 Zillow paid 100k over what a comparable house (26sq ft difference but otherwise same lot, floor plan just with a 2 foot master extension, and limited upgrades, and build year of 2019) sold for in my neighborhood two weeks earlier. They then listed it for +60k over their purchase price. I could understand them snapping up deals but can’t figure out how they can justify overpaying and hoping to still make a healthy profit. My assumption is they are supposed to buy a set number of houses in a month so if they grossly overpay on a few to meet their targets then so be it. （三个月前贴的 早知道就好了 huaren上真的是消息灵通）
https://www.reddit.com/r/technology/comments/p3od64/zillow_other_tech_firms_are_in_an_arms_race_to/ 这里面的讨论也是很开眼的 The consequences are foreseen. People with middle class wages or less will be forced to rent indefinitely. Many will have savings for a down payment but will not be able to close due to being out bid by straight cash. The ones that do manage to get in will be stuck waiving contingencies and likely get wrecked from underlying issues on the overvalued home. It just goes on and on. A buddy just bought, even had a home inspector, but still got walloped by foundation issues. Inspector saw some issues, but didn't seem to think they were major, so my friend desperate for a house pulled the trigger. No clue how he'll afford the foundation issues without a second mortgage, if he can qualify. Your scenario is already playing out for a lot of people. 我记得之前huaren上就有人说过 硬伤房拿出来卖的最佳时机！
zillow shutdown ibuyer
https://www.cnbc.com/2021/11/02/zillow-shares-plunge-after-announcing-it-will-close-home-buying-business.html
Zillow买房翻建竟然血亏
这两天的新闻让我很困惑，Zillow经营亏损不奇怪，哪家公司都可能会遇到，但报道的细节是说Zillow折在购房翻建转卖的业务上。这个业务按理说是稳赚不赔的买卖啊。而且这段时间房价上涨迅速，如果Zillow购房比较早的话，这个房价普遍上涨都会带来很高的利润。这买卖咋还给做亏了呢？
Zillow的说法是折在房价的不可预测性，是否说明当初哄抬房价的就是Zillow？ 说不少买房人看到最后购房成功的都是眼镜不眨现金买入的，难道就是这些人吗？
zillow员工前来回答。
zillow offer的业务ibuying，其实就是以公司的名义，用人工智能算出一个合理的价格，从私人手中收购房屋，然后flip一下再卖出，期待在一个短的周期后去获得小的利润。这个模式的前提是人工智能模型能够精准预测房价，并且没有意外。
但是显而易见事实不是这样。第一，模型用的是历史数据，对于今年的这种通胀没有足够的数据。第二，今年人力成本高昂，材料费翻番，zillow无法在短时间内以原来的价格完成flip。在种种因素的共同作用下，zillow第三个季度巨亏，财报暴雷。
zillow高层遂意识到这个business根本无法scale，而且风险巨大，最终决定关闭zillow offer业务。
我白看着高兴高兴也是好的。
History repeats itself
如果BA是傻逼，DS也只能垃圾进垃圾出。
zillow这公司对自己的股票都没有信心，他家只允许自己的员工每年在一个很小的窗口卖股票，而且卖的话还得给公司写个报告解释为什么要卖，卡的非常严，就是不想让员工卖股票。
我前公司曾经和google某姐妹公司合作过一个接近一亿美金的项目 也是号称用DS啊AI啥的帮我们解决问题。结果就是产生一堆垃圾数据 给我们建了一个垃圾模型。 后来签这个项目的C某O离职了。。
因为很多data scientist分不清overfitting和extrapolation的区别，也不知道该怎么去区别。有些人甚至不愿意承认，完全不考虑extrapolation。一口咬定extrapolation的问题是overfitting。这两者最大的区别是，overfitting是interpolation里的一个问题，是可以通过machine learning/deep learning的各种技术来解决的。但是extrapolatio是技术解决不了的，只能通过收集数据使得samples能cover那个functional space才能解决。但对于high dimensional/nonlinear的问题，这并不容易。一旦承认是extrapolation，可能直接导致项目做不下去了，machine learning/deep learning的作用会大大的限制。而如果不提，不承认，就可以使命的忽悠。但自然世界是不以人的意志为转移，该怎么样就怎么样。这不zillow的人工智能就闹了个大笑话。反正不是建立在严谨的数理基础上的data-driven模型最后都会原形毕露的。能吹得了一时，吹不了一世。
zillow还introduce了一个positive feedback loop
直接造成价格螺旋式上升
嗯，很多data scientist就是这么忽悠的。这药方太熟悉了。
这里面的讨论也是很开眼的
The consequences are foreseen. People with middle class wages or less will be forced to rent indefinitely. Many will have savings for a down payment but will not be able to close due to being out bid by straight cash. The ones that do manage to get in will be stuck waiving contingencies and likely get wrecked from underlying issues on the overvalued home. It just goes on and on.
A buddy just bought, even had a home inspector, but still got walloped by foundation issues. Inspector saw some issues, but didn''t seem to think they were major, so my friend desperate for a house pulled the trigger. No clue how he''ll afford the foundation issues without a second mortgage, if he can qualify. Your scenario is already playing out for a lot of people.
我记得之前huaren上就有人说过 硬伤房拿出来卖的最佳时机！
然后还有这个
Zillow paid 100k over what a comparable house (26sq ft difference but otherwise same lot, floor plan just with a 2 foot master extension, and limited upgrades, and build year of 2019) sold for in my neighborhood two weeks earlier. They then listed it for +60k over their purchase price. I could understand them snapping up deals but can’t figure out how they can justify overpaying and hoping to still make a healthy profit. My assumption is they are supposed to buy a set number of houses in a month so if they grossly overpay on a few to meet their targets then so be it.
（三个月前贴的 早知道就好了 huaren上真的是消息灵通）
extrapolation这事儿基本无解吧，AI就是AI，做不了未卜先知的神仙。
Middle class 去 upper class 地区当然买不起房。全美有大量便宜房子地区。
true！
redfin自己没炒房吧？只是做点中介业务，这个应该稳赚不赔
cathy今天果然是要清仓zillow的节奏了，如果她今天再买，我也要清仓ark了😅
Good points!
同感
这就是典型的光会技术和算法 但是不了解问题本身导致的。以前有很多搞计算生物学的 没在药厂工作过一天 整天叫嚣着挖挖public data就可以研发新药
挂篮子是什么意思？
不需要了解问题本身，现在所谓的ds技术门槛极低，会几行python 调几个人家写好的package最多调调参数生搬硬套一下就号称自己懂人工智能机器学习了
别这样。。我昨天下午买了点，今天卖了，给自己赚了套lego玩
正因为老百姓修房子困难，才是商机啊。但执行的太难看了。修房子再难也没有发射火箭难吧。
本质上所有未来的或者模型没见过的数据不都是extrapolate 么？照这逻辑所有AI ML都不敢承认是extrapolate 要不然都会死很惨？ 我觉得zillow 是不是错在数据质量不好活着real world model drifting 没有及时管控 当然疫情期间人类经济活动和心理都是新的outlier 很难完全捕捉好 reinforcement learning或许可以解决一部分 我不否认对AI 算法过度依赖会产生问题 只是在想怎么加入human loop来更好解决和预防
Zillow这个也是，房地产公司，以及比如连锁餐馆，就不要想着有科技公司疯狂估值了，每一个房子/餐馆的利润都是有限的，一单一单做，如果幻想迅速扩张，肯定只能失败。参见海底捞疯狂扩张也失败了。
我一个朋友在一个公司，也想进入地产行业，要求我朋友的部门每月必须买1000个房子。有这种指标，根本没办法好好筛选，看到房子就买，价高也必须拿下。所以我朋友胆战心惊的觉得他们公司也危险。。。