Read_879 - 阻碍小额支付的心理账户壁垒

微支付系统假设了心理会计问题的解决方案。

A micropayments system assumes a solution to the mental accounting problem.

如果有人能真正解决这个问题，而不仅仅是通过某种神秘手段声称解决了它，智能代理等，即使在长途电话和互联网服务等现有业务中也能节省巨大成本，更不用说所有新的可能性了。

If somebody could actually solve this problem rather than merely claiming to have solved it via some mysterious means, intelligent agents at all, the savings would be enormous even in existing business such as the long distance and Internet service, not to mention all the new possibilities.

比特币领域最棒的内容已登陆Audible。

The best in Bitcoin made Audible.

我是盖伊·斯旺，这里是《比特币可听》。

I am Guy Swan, and this is Bitcoin Audible.

大家好呀？

'Sup, guys?

欢迎回到《比特币可听》。

Welcome back to Bitcoin Audible.

我是盖伊·斯旺，一个比你认识的任何人都阅读更多比特币相关内容的人。

I am Guy Swan, the guy who has read more about Bitcoin than anybody else you know.

本节目由人权基金会赞助，他们通过详实记录和保持与一线人员的联系，为全球人类自由事业做出了卓越贡献。

This show is brought to you by the Human Rights Foundation and their incredible work in, in detailing out and staying in touch on the news and with the people on the ground for the fight for human freedom around the world.

他们的《金融自由报告》是我在该领域见过最优质的资源之一。

Their Financial Freedom Report is one of the best resources I have had on this topic.

你可以订阅通讯，并在节目备注中查看他们其他精彩工作的详细信息。

You can subscribe to the newsletter and check out their other amazing work in the details right in the show notes.

另外，如果你还没读过，亚历克斯·格拉德斯坦的《检查你的财务特权》是一本极好的书，我很荣幸为其做了旁白。

And also, if you have not read it, Alex Gladstein has the, check your financial privilege, which is such a fantastic book, and I was honored to do the narration for it.

如果你想在Audible上获取它，就能听到我用悦耳动人的声音为你朗读这本精彩的书。

So if you wanna grab it on Audible, you will hear my lovely, beautiful voice, reading that amazing book to you.

此外，本期节目也由Bitkit和Synonym赞助播出，他们是PubKey的开发者。

And, this is also brought to you by Bitkit and Synonym, the makers of PubKey.

他们在BitTorrent协议之上构建了一个全新的去中心化DNS系统。

They've built a new system for a decentralized DNS on top of BitTorrent.

我一直在测试这个beta版本，并在Noster之间来回使用它。

And I've been testing out the beta and using it back and forth kinda between Noster.

这真是个既酷炫又充满前景的项目。

It's just a really cool and promising project.

他们开发的另一款产品是BitKit钱包，这是我最喜欢的钱包之一。

And the other the other thing they made is that the BitKit wallet, which is one of my favorite wallets.

这是一款非托管的链上及闪电网络钱包。

It is a non custodial on chain and lightning wallet.

如果你想要一款操作流畅、设计精美且稳定可靠的钱包，若你还没尝试过BitKit，那可就大错特错了。

If you want one that just has a beautiful flow, excellent design, and it just works, If you haven't tried BitKit, you're making a big mistake.

好的。

Alright.

所有赞助商详情都在节目说明里，还包括购买比特币的优质渠道，以及在Fountain上收听和助推本节目的方式。

All the details for our sponsors and everything are right down there in the show notes, and as well as some great places to buy Bitcoin, to listen and boost this on Fountain.

这里有超多优质资源和其他合作方，甚至还有些家庭教育相关内容。

So many great resources and some other affiliates, some homeschooling stuff too.

如果你还没看到这些好东西，我强烈建议你阅读说明。

I highly recommend you read the description if you have not seen all the goodies.

好的。

Alright.

今天，我们实际上要把它带回来。

Today, we are actually bringing it back.

我们拿到了一篇1996年的文章。

We got a piece from 1996.

这是尼克·萨博的一篇旧文，在尼克·萨博之前，我对这个概念思考得并不多。

It's an old Nick Zabo piece, and it's one that I had not thought about this concept a whole lot until Nick Zabo.

其中一篇文章与我们百万年前读过的一篇非常相似，我甚至记不清具体内容了。

One of the pieces this is similar to a piece that we read a million years ago, which I can't even remember the heck.

我记不清确切标题，但这篇与之非常非常相似。

I can't remember the title exactly, but this one's very, very similar to it.

它讲的是心理交易成本。

It's about the mental transaction cost.

探讨微交易的下限是什么，以及涉及哪些动态因素。

Is what is the lower bound on a microtransaction and what are kind of the dynamics involved.

因为这并不像'哦，既然有了闪电网络解决了交易成本问题，我们就能拥有一个成熟庞大的环境'这么简单。

Because it's not so simple as like, oh, we can have this ripe and huge environment of if we just solve the transaction cost problem now that we have Lightning.

我们能否以1聪、10聪、25聪反复交易？能否为所有这些微小产品和服务建立市场？

Oh, can we can we can trade things at one sat and 10 sat and 25 sat and over and over again, we have all these markets for all of these little tiny products and services and goods.

这个体系将因为交易小额便捷且成本低廉而变得健壮庞大。

And and it will be robust and massive because, obviously, it's small and easy and transactions are cheap.

所以现在我们可以做到这些，让所有这些新事物存在。

So now we can do this and and all of this new stuff exists.

但可能存在着比许多人意识到的更大、更根本的障碍。

Well, there might be a bigger and more fundamental barrier to that than a lot of people are recognizing.

我认为在Nostr协议和价值交换及Zaps功能的背景下重新审视这个问题非常有趣，因为我们现在已具备基础设施。

And I think revisiting this is really interesting in the context of Nostr and value for value and Zaps because now we have the infrastructure.

现在我们拥有了网络。

Now we have the network.

现在我们拥有了可以真正重构和重新思考这些盈利模式运作的环境，但我不想忽视那些至关重要的真理。

Now we have the environment in which we can really recreate and rethink about how these monetization models work, and I don't want to lose sight of really important truths.

基于此，我将把话筒交给Zabo。

So with that, I'm gonna let Zabo take it away.

我们将先阅读这篇文章，然后由其他同事进行后续讨论。

We will read this piece, and then we will hit guys take to follow it up.

让我们深入阅读这篇题为《Nick Zabo于1996年撰写的'微支付的心理账户障碍'》的文章。

So let's dive into our article, and it's titled The Mental Accounting Barrier to Micropayments by Nick Zabo, written in 1996.

某些电子商务项目承诺大幅降低交易成本，以实现微支付或微中介等目标。

Some electronic commerce projects promise dramatically lower transaction costs so that we can achieve micropayments or micro intermediation and so forth.

更先进的构想是利用极细粒度的市场来分配计算和网络资源。

An even more advanced idea is the use of very small granularity markets for the allocation of computer and network resources.

这些设想在多大程度上可以实现？

To what extent are such things achievable?

考虑一个与具体支付系统相对独立的特点：费用明细。

Consider a feature fairly independent of the particular payment system: the statement of charges.

这里存在着完整性与复杂性之间的权衡。

Here lies a trade off between completeness and complexity.

一方面，仅汇总收费会为'萨拉米'欺诈创造机会，使分布广泛的虚假或夸大的微收费难以被发现。

On the one hand, merely summarizing charges creates the opportunity for salami frauds, allowing widely distributed false or exaggerated microcharges to go undetected.

此外，仅阅读摘要的各方无法获得反馈来调整其行为以降低成本。

Furthermore, parties reading only the summaries get no feedback by which they can adjust their behavior to minimize costs.

另一方面，过于复杂的声明也让客户难以阅读，导致欺诈、错误和低效使用行为无法被发现，因为一方或双方无法理解收费与服务及付款协议之间的关系。

On the other hand, a statement too complex for customers to read also allows fraud, error, and inefficient usage to go undetected because one or both parties cannot understand the rationale for the charges in relation to the presumed agreement in terms of service and payment.

同样的逻辑适用于在脑海中而非纸上计算这些事项，就像小额现金交易中常见的那样。

The same kind of reasoning applies to working these things out in the head instead of on paper, as is often done in small cash transactions.

市场价格机制运作的基本要求是交易双方都能将费用与获得或提供的价值对应起来，从而相应调整买卖行为。

A basic requirement for market pricing to work is that both sides to a transaction be able to map charges to value obtained or rendered, so that they can adjust their buying or selling behavior accordingly.

这似乎存在一个根本性的认知瓶颈。

There seems to be a fundamental cognitive bottleneck.

对此提出的一个解决方案是智能代理，但由于这些代理是远程编程而非由消费者操作，消费者难以判断代理是在维护自身利益还是对方利益。

One proposed solution to this has been intelligent agents, but since these agents are programmed remotely, not by the consumer, it is difficult for the consumer to determine whether the agent is acting in the consumer's best interests or in the best interest of the counterparty.

可能至少与阅读完整收费声明一样困难。

Perhaps necessarily at least as difficult as reading the corresponding full statement of charges.

此外，让消费者能向代理简单表达复杂偏好的用户界面尚不完善，这可能代表着另一个根本性的认知瓶颈。

Furthermore, the user interface to enable consumers to simply express their sophisticated preferences to an agent is lacking and may represent another fundamental cognitive bottleneck.

通信公司发现计费是主要瓶颈，据估计长途电话费用中高达50%用于计费，全球市场规模约为每年1000亿美元。

Communications companies have found billing to be a major bottleneck, by some estimates up to 50% of the costs of long distance call are for billing, and this is on the order of a $100,000,000,000 per year market worldwide.

互联网服务提供商已转向采用统一费率以最小化这些成本，尽管这会刺激网络资源的过度使用。

Internet providers have been moving to a flat fee in order to minimize these costs, even though this creates the incentive for network resource over usage.

微支付系统需要先解决心理账户问题。

A micropayments system assumes a solution to the mental accounting problem.

如果有人能真正解决这个问题，而非仅通过智能代理等神秘手段宣称已解决，那么不仅在长途电话和互联网服务等现有业务中能节省巨额成本，降低交易成本还将开启无数新可能。

If someone could actually solve this problem rather than merely claiming to have solved it via some mysterious means intelligent agents at all the savings would be enormous even in existing business such as long distance and Internet service, not to mention all of the new possibilities possible by lowering transaction costs.

电费账单。

Electricity bills.

有时账单会以不必要的微小增量记录交易，比如某些电费账单上以100瓦时为最小单位。

Sometimes statements account for transactions in gratuitously small increments, such as the 100 watt hour resolution on some electricity bills.

关于电费账单，大多数人通常不会去深究很多事情，而这些事情本可以让他们从电费支出中获得更多价值。

There are plenty of things most folks normally don't work out regarding their electricity bills, which could improve the value they get for their electricity payments.

哪些电器用电多但个人受益少——这虽然不会显示在电费单上，但可以设想一个连接智能家电的个人记账程序来实现这个功能。

Which appliance are using more electricity with less personal benefit, not available on the electricity bill, but one can conceive of a personal accounting program tied to smart appliances that let you do this.

如何更好地平衡电力供暖与燃气供暖。

How to better balance electric versus gas heat.

你可以详细计算这些来省几块钱，但兼职赚外快来钱更快。

You could compute this in detail and save a few bucks, but you'd earn extra money faster by moonlighting.

如果电力公司是个不太可靠且名声不佳的企业，你可能也不会信任他们的计费方式，而是会按照自己放心的精度重新核算，并接受低于该程度的欺诈或条款陷阱。

If the electricity company was a less reliable and widely known entity, you also might not trust them with the billing and would recompute it to the resolution you felt comfortable with and accept fraud or fine print trickery below that level.

由于电力是可替代商品且定价规则简单，你可以用程序核对账单——如果能帮足够多的人发现条款猫腻以收回软件开发营销成本，这个方式就是高效的。

Since electricity is fungible and the pricing rule set small, you can have a program check the bill, which is efficient if it catches enough fine print shenanigans for enough people to recoup software development and marketing costs.

我们不做这些事的原因是它们不值得耗费脑力。

The reason we don't do these things is that they're not worth the brain cycles.

我们已经触及了心理记账的阈值。

We have reached the mental accounting barrier.

价格粒度理论。

A theory of price granularity.

在此我简要提出一个基于主观主义价格观的价格粒度理论。

Here I present briefly a theory of price granularity based on a subjectivist view of prices.

从购物者的角度来看，价格的功能在于让消费者能够将个人资源（预算）与个人价值相匹配，这些价值是独特且无法直接观察的。

The function of prices from the point of view of a shopper is to let the shopper map his personal resources, budget, to his personal values, unique and not directly observable.

这一心理过程需要将商品购买价格与其个人价值进行比较。

This mental process requires comparison of the purchase price of a good to its personal value.

这意味着显著的心理成本，这构成了交易成本最基本的底线。

This entails a significant mental cost, which sets the most basic lower bounds on transaction costs.

例如，比较大量低价商品的个人价值可能需要耗费的心理支出超过这些商品本身的价格——这种心理支出可以量化为：因未将脑力劳动用于赚取工资而产生的机会成本，或未选择购买数量较少但更具可比性、心理核算成本更低的商品所造成的机会成本。

For example, comparing the personal value of a large diverse set of low priced goods might require a mental expenditure greater than the prices of those goods, where mental expenditure may be measurable as the opportunity costs of not engaging in mental labor for wages or of not shopping for a fewer number of more comparable goods with lower mental accounting costs.

在这种情况下，将商品打包成价格更高、具有直观协同效应的组合套餐，直到购物者的心理核算成本足够低，才是明智之举。

In this case, it makes sense to put the goods together into bundles with a higher price and an intuitive synergy until the mental accounting costs of shoppers are sufficiently low.

决定价格粒度主要下限的是这些心理核算成本，而非支付或计费方式的物理成本、计算成本或研发分摊成本。

These mental accounting costs, not the physical or computational or amortized R and D costs of payment or billing method, set the main lower bound on price granularity.

从定价粒度趋势来看（比如在线服务普遍采用统一费率），网络定价粒度远高于建议的微支付水平（几美分甚至更小面额）。

Judging from pricing granularity trends, such as the trend towards flat rates in online services, online pricing granularity is far above suggested micropayment levels of a few cent or even fractions of a cent.

典型网络消费者的心理核算成本似乎比传统熟悉商业领域的成本略高。

The mental accounting costs for a typical online consumer seem to be somewhat higher than those in more familiar areas of commerce.

一个可能的解决方案是让购物软件通过消费者报告服务来比对购买价格，但这类客观信息很稀缺，且无论如何只能反映普遍共性价值而非个人价值。

One possible fix is for the shopper's software to compare purchase prices against a consumer reports service, but such unbiased information is rare and in any case takes into account only widely shared values, not personal values.

对于可替代商品还有另一种解决方案：按单位收取固定价格，消费者仅需根据累计数量和价格信息即可评估。

Another fix is possible for fungible commodities: charge a fixed price per unit, which the shopper can evaluate from just the accumulated number of units and price information.

具体案例：在美国当前广告战中，AT&T押注其15美分统一费率比Sprint那说不清道不明的10美分浮动费率更具吸引力——值得供应商放弃拥堵定价，消费者放弃深度折扣，以换取可预测费率，将通话时长转化为可替代商品从而节省心理核算成本。

As a concrete example, in a current US ad campaign, AT and T is betting that its 15¢ flat rate is more attractive than Sprint's 10¢ to who knows what variable rate, that it is worth the vendor forgoing congestion pricing and the shopper forgoing deep discounts in order to have a predictable rate, turning phone time into a fungible commodity and thus saving on mental accounting costs.

可惜大多数互联网商务内容（服务、邮购商品等）都不可替代。

Alas, most Internet commerce is not fungible content, services, mail order products, and so on.

某些互联网服务提供商（ISP）的服务可作为可替代商品出售，例如磁盘空间或连接时长，但这需要放弃拥堵定价等定价策略——若非心理账户成本的存在，这些策略本可能非常高效。

Some Internet service provider or ISP services can be sold as fungible, for example disk space or connection time, only at the expense of foregoing congestion pricing and other pricing methods that, if it were not for mental accounting costs, might be quite efficient.

此外，即便对于可替代商品，每个用户也都有独特的收益递减曲线。

Furthermore, even for fungible commodities, each user has a unique curve of diminishing returns.

软件需要让购物者以某种直观熟悉的方式确定并输入其数量偏好曲线，且不能预设购物者熟悉经济理论，才能充分代表其利益——这还未考虑时间偏好、孤立状态下可替代商品间的非线性交互等复杂因素。

Software would have to let the shopper determine and input his volume preference curve in some intuitively familiar way without presupposing the shopper is familiar with economic theory before it could adequately act in his interests, not to mention the complications of temporal preferences, nonlinear interactions between commodities fungible when in isolation, and so on.

针对可替代商品的这一用户界面解决方案，为攻克线上消费中更普遍的心理账户问题提供了更好策略：开发更有效的方式让购物者向软件传递个人偏好。

This user interface solution for the case of fungible commodities suggests a better strategy for tackling the more general problem of mental accounting in online commerce: develop better ways for the shopper to communicate his personal preferences to software.

营销人员长期设计各种方案获取此类信息：详细问卷、用户行为与反馈追踪等。

Marketers have long devised schemes to get this kind of information: detailed surveys, tracking of user behaviour and responses, etc.

可以说，像www.firefly.com这类网络服务在此领域最为先进。

Arguably, web services like www.firefly.com are the most advanced in this regard.

Firefly构建了一个音乐偏好的主观空间，购物者可在此探索并发现更可能喜爱的新音乐。

Firefly creates a kind of subjective space of musical preferences in which the shopper can navigate and find new music that they are more likely to prefer.

既然软件能表征特定偏好，那么将这些表征映射到具体价格、完成购物并安全执行在线交易对软件而言是更直接的问题。

Given that software can represent certain preferences, it is a more straightforward problem for software to map these representations to specific prices engage in shopping and securely complete online transactions.

这些较简单的问题一直是微支付研究的焦点，但获取并表征偏好这个更基础的问题却大多被忽视，或许源于将数学法则而非主观偏好视为经济运作基础的客观主义偏见。

These easier problems have been the focus of micropayments research, but the more fundamental problem of obtaining and representing preferences in the first place have gone largely unrecognized, perhaps due to an objectivist bias that posits mathematical laws rather than subjective preferences as the basis of a working economy.

在其他交易成本问题解决后，受偏好传递过程限制的心理账户成本问题便凸显：购物者A能在多大程度上以何种效率向软件传递主观偏好？软件B能否表征并基于这些具象化偏好行事？

Given the solution of other transaction cost problems, mental accounting costs that become subject to the limit on the process of communicating preferences To what extent and with what efficiency can A, a shopper communicate subjective preferences to software and B, can software represent and act in the interests of these objectified preferences?

搜索引擎、商品目录订单、营销问卷以及Firefly这类复杂交互的存在，证明这种沟通与表征既可行又重要，但似乎成本高昂且可能存在某种根本性限制。

The presence of search engines, catalog order forms, marketing surveys, and more sophisticated interactions like Firefly demonstrate that such communication and representation is both feasible and important, but seems to be costly and perhaps fundamentally limited in some way.

偏好与视觉隐喻。

Preferences and visual metaphors.

为评估交易的可取性并避免被多收费，交易双方需核算为特定商品和服务支付的款项，无论是确保现金支付如约履行或后续收据核对，还是确认电话账单无误。虽然我用现金支付，但仍希望追踪资金流向及原因，这与会计核对分析账目的诸多理由相同。

To assess the desirability of a transaction and to avoid being mischarged, the parties to a transaction have to count up the money paid for particular products and services, whether making sure that cash payments are made as promised or the receipt afterwards, or making sure the phone bill is proper I may be paying in cash, but I'd still like to keep track of how and why my cash is going in and out, for many of the same reasons that accountants reconcile and analyze book entries.

目前，无论是电子现金还是信用卡的交易记录，都是实现这一目标最有效的方式。

Right now, a transaction log, whether eCash's or aCredit cards, is the most useful way to do this.

某些情况下可能存在更贴切的比喻，例如绝对水平计量器、带有高低水位的速率计量器等。

There may be other metaphors more appropriate for some circumstances, for example absolute level gauges, rate gauges with high and low watermarks, etc.

这是一个值得探索的潜在新兴领域。

This is a potentially fertile new field to explore.

可能存在能处理部分会计工作的代理程序，例如将实际支付与承诺条款、支付限额等进行比对。

There may be agents that can do some of the accounting, for example comparing payments made to terms promised, payment limits, etc.

但对于绝大多数商品和服务，软件无法判断其质量或个人需求，因而也无法评估交易的净可取性。

But for the vast majority of products and services, software cannot judge the quality or personal desire for the product or service, and thus the net desirability of the transaction.

用户必须基于计算机通过界面提供的所有信息自行完成这种评估。

The user must undertake this comparison with whatever information the computer can provide via the display.

因此用户界面和用户认知能力仍是制约交易颗粒度的瓶颈。

The user interface and the cognition of the user thus remain the bottleneck to transaction granularity.

重要课题是利用图形用户界面的力量创造新比喻来简化这一过程。

A big task is to use the power of the graphical user interface to come up with new metaphors to make this easier.

正是直观又精确的比喻才能降低会计成本。

It is the intuitive yet accurate metaphor that will lower accounting costs.

加密协议主要降低的是与安全相关的交易成本，例如伪造和勒索。

Cryptographic protocols potentially lower only security related transaction costs, such as forgery and extortion.

对于当前阻碍小额支付的常规会计交易成本，我们需要更好的交互式视觉比喻。

For the normal accounting transaction costs, which are currently too high for micropayments, we need better interactive visual metaphors.

要实现无记录交易，我们需要能公平执行一次的交易，通过巧妙的视觉隐喻让双方即时确认，之后便可遗忘。

For transactions free of records, we need transactions that can be fairly transacted once, immediately accounted for by the parties, via a nice visual metaphor, then forgotten.

在无记录系统中，对于无法实现即时确认的交易（很可能是大多数理想商业场景），存在大量无法解决的纠纷风险，比如产品或服务质量需在交易完成后才能判断，或涉及信用交易的情况。

The potential for unresolvable disputes in record free systems is vast for transactions where this is not possible, probably most of desired commerce, where quality of a product or service cannot be well determined until after the purchase transaction is complete, or where credit is involved.

价格是合同条款的一种形式。

Price is one kind of contractual term.

我们还需要巧妙的隐喻来追踪其他类型的合同条款。

We also need nice metaphors to keep track of the other kinds of contractual terms.

当用户无法观察协议执行时，交易对手方就可能采取隐蔽行动。

Lack of observability of the protocol on the part of the user leads to the ability of the counterparty to engage in hidden actions.

关于这个议题及其他计算机化合约问题的深入讨论，请参阅《数字市场构建模块》。

See Building Blocks for Digital Markets for further discussion of this and other computerized contracting issues.

制定优质合同的障碍之一，首先在于确定各方真正想要什么。

One of the barriers to creating good contracts is determining what the parties want in the first place.

人们往往局限于标准化的思维模式：用美元支付、投资股票等等。

People tend to think in terms of standard or stereotyped conditions: payment in dollars, investing in stocks, etc.

实际上存在更多样化的合约结构选择，若能合理组合，将能更好地满足各方需求。

When there exists a far wider variety of alternative contractual structures that, combined properly, could better meet the parties' needs.

我希望开发出能让各方通过计算机交互式探索需求的工具。

I'd like to see tools which allow parties to explore their desires interactively with the computer.

在金融领域，这可能包括交互式个人收益率曲线分析，像决策理论那样确定需求偏序关系，针对特定替代证券、衍生品和合成产品等进行设计。

In finance, this might include interactive personal yield curves, determining the partial order of desires, as in decision theory, for particular alternate securities, derivatives and synthetics, and so on.

软件随后会分析这些输入，提出建议，甚至执行自动化合约签订。

Software would then analyze this input, make recommendations, and even undertake automated contracting.

应当开发隐喻手法，让非专业用户无需掌握大量金融或决策理论就能轻松表达这类需求。

Metaphors should be developed to make it easy for lay users to express such desires without extensive knowledge of finance or decision theory.

这类隐喻将为自动化交易、拍卖及其他在线合约机制提供友好的前端界面。

Such metaphors would provide a friendly front end to automated exchanges, auctions, and other online contracting mechanisms.

目前像Quicken这样的预算程序提供部分隐喻功能，金融分析程序能对特定合约的现金流特性提供详尽反馈，但将这两种技术结合还存在巨大的未开发市场潜力。

Currently budget programs like Quicken provide some of the metaphors, and financial analysis programs provide extensive feedback on the cash flow properties of particular contracts, but a potentially large, untapped market lies between in a combination of these two technologies.

好的。

Alright.

我想重提这个话题有几个原因——虽然节目里没讨论过，但Nick Zabo曾就心理交易成本的概念做过类似探讨。

So there's a couple of different reasons why I wanted to bring this, piece back, which we haven't covered it on the show, but there's, again, there's one that Nick Zabo did similar to this one on the whole idea of mental transaction costs.

但随着Noster的普及环境形成，我们既看清了微支付的本质，也见证了Fountain对流式支付等概念的探索。

But with the adoption and the environment of Noster and kind of seeing what micropayments actually are, and then also seeing streaming sets and that sort of thing explored in Fountain.

我认为重新审视比特币在这些环境中的成功案例与现存障碍是很有价值的实践。

I think it's a useful exercise to go back through and think about what has succeeded and or has still been a barrier for those types of environments and those types of monetary interactions with Bitcoin.

同时在设计新盈利模式时，既不应过度执着，也要认清看似微小支付背后的实际障碍——因为真正的微支付并不存在。

And then also in thinking about new monetization models is not getting overly attached to or failing to recognize the barriers to to the seemingly tiny payments because there is no micropayment really.

正如Zabo所说：下限取决于个人价值体系中与其他选项的对比能力，这本身就存在心理障碍。

Because as Zabo says, the lower bound is the ability to compare that payment according to personal values against alternatives, which has a mental barrier.

这种心理成本限制了任何交易能达到的精细度和最小单位。

It has a mental cost that limits how granular and how small any transaction can actually be.

用他举的例子来说：假如有100种价格略有差异的相似微商品供选择时——

So using the example that he talked about bringing up is just, well, what if there's, like, a 100 different options of very similar but varied and slightly different priced, like, micro goods.

对吧？

Right?

嗯，如果我们讨论的是一分钱、两毛钱，甚至是一块钱的东西。

Well, if we're talking about something that's a penny, a 20¢, even a dollar, really.

而且，你知道，我们就像是在查阅评论，试图比较这个东西的各项参数。

And, you know, we're, like, going off and reading reviews and trying to trying to compare, like, what are the stats on this thing.

就像我在看这台电脑有多少内存和CPU之类的。

Like, I'm looking at, you know, how much RAM and CPU does this computer have kind of thing.

就像我在购物。

Like, I'm I'm shopping.

购物是一个过程。

Shopping is a process.

购物实际上是一种相当复杂的评估过程，用来判断你将获得的产品信息。

Shopping is a is a kind of sophisticated process actually for making an assessment of what the information is about the product that you're going to be receiving.

比如，你可能是个糟糕的购物者。

Like, you can be a crappy shopper.

你知道，有些人看着两台电视会说'这台只有10AP而那台是2K'。

You know, you can have someone who is looking at two TVs and says, oh, well, this one's only 10 AP while this one's two k.

他们根本不知道这两者其实是相同的。

And they just don't know that that's the same thing.

这只是1920×1080的不同营销术语——不知为何当我们翻倍时就改称4K了。

It's just a different branding for for, you know, 1920 by October because for some reason, we just switched to four k when we doubled that.

可能在广告里听起来更吸引人吧。

Sounds better in the ad, I guess.

而且这也像是一种心理账户障碍——好吧。

And it's also just you kinda have a mental accounting barriers like, okay.

那么，这个与上一个有多少关联呢？

Well, how many how many how does this relate to the last one?

四k代表一个数字和一个字母。

Four k is one number and one letter.

而且这似乎是个不错的副作用——它的效果是1080p的四倍。

And it's just kind of a nice side effect that it seems to be four times as good as ten eighty p.

要真正达成好交易需要大量技巧和知识，因为你不仅要了解个人品味、价值观和偏好，还要理解交易条款。

There is a lot of skill and knowledge required to actually get a good deal because you have to know what the not only what your personal taste and values and preferences are, but what the terms of the deal understand.

你看，这就像买保险合约一样，只是细节和合约条款的透明度有所不同。

You know, you think about it like everything is basically like buying an insurance contract with just the the various the variant in details and visibility into what the contract means.

比如电视合约你至少能看清大部分条款，但保险合约、付款计划或利率的细节就看不透。

Like, the contract that you have with the TV, you feel like you can at least see most of it, but you can't with the details of an insurance contract or a payment plan or interest rate.

大多数人完全不懂利息，这导致他们在融资或借贷时非常吃亏。

And most people don't understand interest at all, which makes them terrible shoppers when it comes to financing something or taking out debt.

即便你拥有技术方案来解决某些细粒度的支付问题，比如按分钟计费听内容或购买各种小物件。

So just because you have a technical solution to some sort of, you know, granular payment problem or pay per minute to listen to something or buy all these tiny little things.

当需要做比较时，如果商品不具备可替代性——就像按API调用计费那样，你购买的是像扎瓦举例的电费账单那样，用明确金额换取特定商品。

Like, it gets to a point where if if you're having to make comparison, if it's not some sort of fungible thing, like, that's that's the good thing about, like, a pay per per API call or something like that is that you could have a fungible item that you're purchasing using the electricity bill that Zava brought up as the example, is you have some very specific thing that you are purchasing per some very explicit amount.

每次交互时购买的都是完全相同的标的物。

And every single time you interact with it, it's the exact same thing.

就像流媒体统计或按分钟计费这类服务。

That's like streaming stats or, you know, pay per minute, whatever it is.

但也要注意，多数人特别喜欢——比如AI推理就是个很有趣的例子。

But also notice, most people like to especially when it comes to like, AI inference is a really interesting one.

因为你无法确切知道会得到多少令牌。

Because you don't know exactly how many tokens you're gonna get.

比如，会有多少单词、段落，或者进行多深入的研究。

Like, how many words or paragraphs or how much deep research is going to occur.

所以如果你只是为计算付费，我怎么知道我即将回答的问题会产生20美分的答案还是7美元的答案呢？

So if you're just paying for computation, how do I have any idea that the question I'm about to answer is either going to produce, you know, a 20¢ answer or a $7 answer?

这正是订阅模式如此受欢迎的原因。

Which is exactly why the subscription model is so unbelievably popular.

它消除了这些心理交易成本。

It takes away those mental transaction costs.

但我认为它也被应用到了许多本不需要订阅模式的地方，由于人们不擅长这种心理计算，它被过度滥用了。

But I also think it's being applied to a bunch of things that don't need a subscription model, and it's been aggressively overused because they're trying to take advantage of people not being good at that mental accounting.

然而有趣的是，我在Noster社交平台上随意浏览时没有这个问题。

However, what I find interesting is that I don't have this problem with the social side of Noster when I'm just zapping.

我认为这很合理，因为我不需要去比较不同的社交互动。

And I think it makes sense just because my I don't have to think about comparing my social interactions.

比如，我不需要去比较100种价格在1便士到50便士之间的微小差异，我根本不会花时间在这上面。

Like, I don't have to you know, if I'm I'm trying to think of, like, know, going back to the idea of the comparing a 100 different variant tiny things that are between 1p and, like, 50¢, like, I'm going I'm not gonna spend any time on that.

如果我不了解那些条款的含义，我甚至懒得去购物或弄清楚它们。

Like, I'm not even gonna bother trying to shop or figure out what all the terms mean if I don't know that.

我连看都不会看。

I'm not even gonna read them.

如果我现在急需解决某个问题或立即使用某物，我可能就直接买来用了。

I might just if I'm trying to get something to solve some sort of a problem or that I can interact with right now, I'm just gonna kinda buy something and use that.

如果它不工作，那就不工作吧，我可能会再买一个。

And if it doesn't work, then doesn't work, I'll maybe get another one.

我不会只是可能离开这件事，而是会完全放弃，寻找某种已经包含所需功能的平台。

I just won't just maybe I'll just walk away from this whole thing entirely and look for some sort of a platform that has the thing included.

我认为很多这种想法的问题在于，当你面对一个如此微不足道以至于显得可笑的事情时，因为试图定价的人也同样面临这个问题。

And that's really where I think a lot of this sort of thinking goes, is when you have something that's so small as to make it silly, because the person trying to price the thing is also having the same problem.

你如何定价才能让用户不会觉得‘这是什么玩意儿’？

How do you how do you price something in such a way that the user doesn't just like, what is this?

为什么我...我甚至不想讨论这个。

Why am I I don't even want to go here.

你知道，这就像在登录或创建个人资料时额外增加两个步骤。

You know, it's like adding two extra steps into, like, logging in or creating a profile or something.

这只会惹恼人们。

It just annoys people.

当你做某件事时，你是为了获得某种好处，或是为了某种体验、社交环境或娱乐，你最不想要的就是一堆文书工作，那种精神上的文书工作，以及需要做出决策的压力。

When you're doing something, you're you're getting some sort of a benefit or you're you're going somewhere for some experience or social environment or entertainment, the last thing you want is a bunch of paperwork kind of, you know, the the kind of like mental paperwork and the idea of trying to like stress out or make decisions.

很多人会做一些事情来放松或解决特定问题，就是为了避免做出一大堆复杂或恼人的决定。

Like a lot of people will do things to relax or to solve some sort of problem specifically to avoid having to make whole bunch of complicated or annoying decisions.

所以这就是为什么一堆小功能或附加项会被随意加进来。

So that's why a bunch of tiny little features or additions just kinda get added in.

你看那些订阅服务，比如播客托管服务之类的在线服务，免费版或低价版通常只包含核心功能。

You know, when you look at like a a subscription service or something that's, know, like a podcast hosting service or whatever it is online, you'll get the the free version or the low cost version that just kinda has like the core.

比如，这三样东西才是你真正想要的。

Like, these are the three things that you're really kind of looking for.

然后下一个等级的价格大概要贵一倍。

And then the next tier up is, like, twice as expensive.

比如说第一个等级是10美元，下一个是20美元，而那个所谓的'热门'选项就在列表里。

Let's say the first one's, like, $10, the next one's $20, and it's the quote, unquote popular one when you when you see it in the list.

但他们基本上就是把所有那些额外的小功能都加进去，你根本不知道该怎么精确收费。

But they basically add all of those extra little features that you would never know what to charge exactly for.

就像，哦，我举个播客托管的例子吧。

It's like, oh, I'm gonna have a go back to podcast hosting as an example.

我会加上转录功能，然后它会自动生成章节。

I'll add in transcriptions, and then it'll make chapters for you.

你还可以有动态广告插入之类的功能。

And you can have dynamic ad insertion or whatever it is.

然后就像有个项目符号列表，列出了你能得到的另外10个功能。

And there's just like this bullet point list of, like, these 10 other things that you get.

所以你只需要比较这些功能的总体情况和你可能用到的部分，对比最基本的'我只需要一个地方存放播客'的需求。

And so all you have to do is compare the general overview of what those things are and what you think you'll use versus the the basic, like, oh, I just need a place to store my podcast.

这些功能没有被单独列出来是有原因的，不然你可能会看到'转录99美分'、'自动文件转换1.5美元'或者'自定义图片每张20美分'这样的收费。

There's a reason why all of those things aren't itemized out, where you don't add 99¢ for transcription or a buck 50 for, auto file conversion or 20¢ in order to check do custom images for every single thing.

那样会让人抓狂的，而且说不定还能更高效些。

Like, it would just be infuriating, and you might even be able to get it more efficient.

你可能会想细化收费项目。

You might be able to granularize.

比如，也许我不想每期节目都换图片。

Like, maybe I don't wanna change the image every single time I do an episode.

我不需要那个。

I don't need that.

这里我能省下20美分。

I can save 20¢ here.

但现在我立刻就想构建这个，像是那种精心设计的、定制化的配置组合，而我甚至都不确定自己百分百需要它。

But now I'm immediately trying to build this, like, big elaborate kind of, like, custom configuration set that I'm not even a 100 sure I'm going to need.

所以你知道，三周后我可能会意识到，哦，其实我想换掉这张图片，因为，我不知道，可能节目里会来个大人物。

So, you know, in three weeks, I might realize that, oh, I actually would like to change the the image of this one because, I don't know, somebody somebody really big is on the show.

戴夫·史密斯要上节目了。

Dave Smith is gonna be on the show.

所以，就像，哦，那把他放到图片上吧，这样人们就能在推送里看到戴夫·史密斯要上节目了。

So, like, oh, well, let's put him on the image so that people can see that Dave Smith is on the show in the feed.

现在我又得回去给我的方案增加20美分，还要修改配置细节。

Well, now I gotta go back and add 20¢ to my plan and change the configuration detail.

巴拉巴拉。

Blah blah blah.

对吧？

Right?

我现在就是被一堆本身没那么重要的事情压得喘不过气。

I just now there's this huge ridiculous burden about a bunch of things that don't matter that much themselves.

我发现这篇文章另一个有趣的点是他提出了代理人的概念。

Now another thing I found interesting about this piece is that he brings up the idea of agents.

这是1996年的事。

So this is 1996.

当时还没有AI代理。

There were no AI agents.

对吧？

Right?

那时候还没有聊天机器人。

There was no chatbot back there back then.

他只是在考虑那种可能具有某种简单算法来做评估的软件。

He's just thinking of, like, software that can potentially have some sort of a simple algorithm for making that assessment.

我觉得很有趣的是算法基本上替我们完成了这些，但这些都是中心化算法，它们试图把认为我们会购买的东西推到我们面前，或是推送那些会让我们——你知道的——被愤怒情绪吸引而留在推特或其他平台上的互动内容。

And I find it kind of interesting that algorithms have basically done this for us, but they've been centralized algorithms that have tried to push stuff in front of us that they think we're going to buy or push interactions which are gonna keep us, you know, rage baited to stay on Twitter or whatever it is.

所以这些代理已经被广泛使用。

So these agents have largely been in use.

他们以Zabo使用的Firefly为例，比如音乐推荐系统，现在显然各种平台都有这种功能。

They use Zabo used a a Firefly as the example of, like, a music recommendation, which obviously you now get in all sorts of thing.

Netflix会根据你看的内容、看的预告片、完整看完还是中途停止、每集观看时长来给你推荐。

You get Netflix recommendations based on what you watch, based on what the trailers you watch, and which ones you you continue you watch the whole thing versus stop it versus how much you watch of each episode.

它们会预测你可能喜欢哪些内容，不喜欢哪些内容。

They give you a prediction on which one you're probably gonna like and which one you won't.

亚马逊商品和线上购物也是同样的道理，很多这类系统会通过社交媒体追踪你，在互联网上跟踪你访问的网站，然后把相关商品推送到你面前。

Same thing with Amazon Amazon products and online and, you know, a lot of these things track you across your social media and attract track you across the Internet to see what websites you go to, and then they stick those items in front of you.

这非常隐秘，虽然大家都矢口否认，但巧合实在太多了。

And there is a very sneaking and kind of everybody says it's not true, but the coincidences just stack.

这是个好笑的现象，因为所有人都有过这样的经历：在谈话中随口提到某样东西，接着就在使用手机时看到亚马逊或其他平台推送相关广告或推荐，大家都觉得自己的对话被录音了。

The the fact that it's a funny joke because everybody has had the experience of saying something out loud during the conversation and then getting an advertisement or getting it recommended to them on Amazon or some other environment when they're using their phone next, everybody feels like something's being recorded.

比如，他们的手机正在监听他们，并根据他们的环境和想要购买的东西或说的话进行适配。

Like, their phone is listening to them and adapting to their environment and the purchases that they wanna make or the things they say.

哦，他们正在谈论要去度假。

Oh, they're talking about they're talking about going on a vacation.

多么巧合啊，我刚收到Airbnb关于去山区的新通知。

And how what a coincidence that I just got a new notification from Airbnb about going to the mountains.

就像很多人说的那样，哦，这不可能。

It's like a lot of people are like, oh, that's not happening.

这不可能。

That's not happening.

但大家都因为这个笑话而笑是有原因的。

But there's a reason why everybody laughs at that joke.

每个人都知道那种感觉。

Everybody knows that feeling.

每个人都知道那种感觉，手机和那些公司比你更了解你自己，它们在监听并收集每一条小信息。

Everybody knows that sense that the phone and those corporations know you better than you know yourself, and that they are listening and taking in every little piece of information.

今天在听的你们每个人都知道。

And everyone of you listening today knows.

你们都有这样的经历，是的，我们在私人谈话中提到了某件事，然后突然，我们中的一个人就收到了相关的广告，或者它出现在电子邮件通讯中，诸如此类。

You you have that story of, yes, we were talking about this thing in a personal conversation, and then boom, one of us got an ad for it, or it showed up in an email newsletter, whatever.

这某种程度上就是代理概念的体现。

That's kind of where the, the idea of the agent has actually manifested.

但在人工智能时代，有趣的是这种情况可能会改变。

But in the age of AI, it's actually kind of interesting that that this could change.

但我并不确定这真的能解决问题。

But I'm not so sure I'm not so sure it really solves the problem.

我觉得很多人固执地坚信AI会孕育出超级智能之类的，但我还是不太确定——你知道有篇很棒的文章，是讲未来十年的？

I think a lot of people are hell bent and, like, just totally convinced that AI is, like, gonna we're gonna birth super intelligence or whatever, and I still don't you know, there was a great the great piece by what was the one about the next ten years?

哦，是Ashenbrenner写的。

Oh, Ashenbrenner.

Ashenbrenner是作者的名字。

Ashenbrenner was the the author's name.

我记不清了，但在《AI Unchained》里能找到，你可以在这个订阅源里看看。

I can't remember, but it's on AI Unchained, which you can find in this feed.

我会试着记起名字并把它找出来。

I'll try to remember the name of it and and dig it up.

但我建议你听听那个，因为我反复思考过超级智能是否会被创造出来，以及更大的模型是否能创造出更高效的小模型——如果你叠加或堆叠训练规模，就能让模型变得越来越好、越来越大。

But I I recommend you listen to that because I went back and forth quite a bit about whether or not super intelligence was going to be created and if the logic of a bigger model can create a smaller model that's more effective, and that can lead to if you then compound or or stack, you scale the training of that model, then that can make it better and better and better, bigger and bigger.

所以成功的巨型模型实际上能训练出更大更好的模型，而不需要自身变得更聪明才能创造出更聪明的模型。

And so big models, big successful models can actually train bigger and better models without having to be or understand, like, be smarter themselves in order to make a smarter one.

因为用同样的方法，你可以造出大模型，也可以造出小模型。

Because being able to make a small you you can make a from the same recipe, you can make a big model or you can make a small model.

如果你造出小模型，可以继续增加算力来把它变成大模型。

And if you make a small model, you could continue to add compute to that in order to make a big one.

既然能做到这点，就意味着你可以造出两个小模型。

And so if you can do that, that means you can make two small models.

由于它必然比创造它的大模型'笨'，那么它就能做决定——基本上可以测试所有模型，然后说：'右边这个模型，让我稍微调整下方法，重新训练或做些反馈学习，让它取得更好结果。'

And because it's necessarily smaller, quote, unquote, dumber than the big model that's making them, well, then it can decide, it can basically test them all and say, like, okay, well, this model on the right, let me change let me tweak the recipe a little bit, let me tweak the how we do this a little bit and retrain it or do some feedback learning, whatever it is, to get it to have better results.

现在你有两个不同的小模型，它们基本上都可以作为大模型的基础配方，其中一个现在比另一个更好，因为它们都比大模型更简单。

And now you have two different small models that both could just basically be fundamental recipes for a big model, and one of them has now been made better than the other because they're both dumber than the big model.

然而，如果你采用那个更好模型的配方，那么那个不太好的第一个模型的配方实际上就是评估者——那个大模型本身的配方。

However, if you then take the recipe of the better one, the recipe of the first one that's not as good is the recipe for the model itself, the big one that's making the assessment.

现在如果它能改进小版本，就能通过计算将其发展成大模型，而这个新大模型会比原始大模型更好，因为它完全是从小模型衍生而来的。

And now if it can make the the small version better, then it can compute and grow it into a big model, and now the big model is better than the original big model because it's simply derived from the little one.

这就是背后的逻辑，也是为什么会有'AI可以训练出更好的AI'这种观点。

So that's kind of the logic behind it and why there is this idea that AI can train better AI.

但我并不...这篇文章提出了非常非常有说服力的论点，但我越来越不确定了。

But I am not so and it's a very the piece was a very, very convincing argument, but I'm increasingly not so sure.

我越来越不确定了。

I'm increasingly not so sure.

我认为这其中存在几个问题，比如Zavo在文中指出的：这一切都是主观的。

I think there's a couple of different problems with that is well, one of the things that Zavo brings up in this piece is that it's all subjective.

在某种程度上，这完全是主观的。

At some point, it is all subjective.

在AI模型的准确性和价值判断中，客观性是有限的，因为我们选择某个而非另一个的许多决策和理由都是明确主观的，这意味着你需要有一套价值观和偏好体系，而AI并不具备这些，因为这不是AI的本质。

There is only so much objectivity in the accuracy and in the value judgment of an AI model because so many of our decisions and reasons to pick one over another are explicitly subjective reasons, which means you have to have a set of values and a set of preferences, and AI doesn't have that because that's not what AI is.

AI只有概率。

AI just has probabilities.

仅此而已。

That is it.

还存在另一个问题：即使你有一个小模型（虽然通常确实有效），但并不能仅因为你能让一个小模型比另一个小模型更好，就100%确定大模型也会...这其中有很多细微差别，在创建这些模型时会出现很多意外情况。

Then there's another problem is that just because you have a small model, like, it does tend to work that way, but it is not certain that just because you can make a smaller model better than another smaller model, that the bigger model will a 100% there's a there's a lot of nuance, and there's a lot of unexpected things in apparently creating these models.

某种程度上来说，这有点像图像扩散——你使用随机种子，但有些生成的图像可能效果很差。

There's there's a there's a big degree of it's kinda like image diffusion is you use random seeds and some of your images might just suck.

显然，根据我从人们那里听到的讨论和创建模型过程中遇到的挑战来看，模型创建本身也具有一定的艺术性，这些都深深体现了中本聪的精神。

Well, apparently, there's this there's a bit of an art to creating models as well from what I've heard in talking to people about it and in the challenges of creating them shared a lot of this on the spirit of Satoshi.

但正因为如此，你其实并不清楚。

But because of that, you don't really know.

如果你没有评估方法，无法衡量或判断该模型的输出，那么你就无法100%确定它比另一个模型更好。

If you don't have a way to assess it, you don't have a way to value or judge the outputs of that model, then you don't a 100% know that it's better than the other one.

随着规模不断扩大，计算需求持续增加，成本会越来越高，我认为边际收益将会递减。

And the bigger and bigger you compound, the the more you scale the commute compute, the higher the cost, and I believe we're gonna have a a lower marginal return.

我们增加的计算资源越多，边际收益就会越低，这主要有两个原因。

The more and more compute we add, the lower and lower the marginal return is for two different reasons.

其中一个原因与整篇文章中提到的心理交易成本问题密切相关。

And one of them is deeply related to the whole mental transaction cost problem, in this piece.

第一个原因是：在小规模模型能力提升过程中，实际能获得多少容量和性能变化。

The first reason is that how much actual capacity and variation can you get in the capabilities of small models that you can scale up.

而当你进行扩展时，其规模会呈指数级增长。

And when you are scaling up, it scales up exponentially.

这些大型模型对计算资源的需求变得极其疯狂——需要大量计算资源才能将它们保持在内存中运行，使其具备可用性。

So you're needing bigger like, these big models get really insane with the degree to which they are they need compute in order to keep them in RAM, to keep them in memory, to make them viable.

相比之下，我们已经使用了大量其他机制来创建小型模型或专用模型，并让它们相互协作——我认为这总体上会是更好的方法。

Whereas, we're already using a ton of other mechanisms to make smaller models or make specialized models and have them interact with each other, which I think is going to be the far better approach overall.

未来不会只有一个庞大到荒谬的单一模型。

It's not going to be just one big giant ridiculous model.

但另一个正处于评判状态。

But then the other one is in judgment.

关键在于你如何知道这个比那个更好？

Is in how do you know this one is better than the other one?

更重要的是，如果它的计算成本是两倍，也就是说，如果你让我为使用这个模型支付两倍的费用，就像是在追求超级智能的概念。

And importantly, if it costs twice as much in compute, like, if you're if you're charging me twice as much to use this model as going to the idea of super intelligence.

如果你问我是否愿意为超级智能支付十倍的费用，而只为见过的最聪明人类支付10%的费用。

If you're asking if you're asking me to pay 10 times as much for the one that's super intelligent versus pay 10% to the one that's just, like, the smartest human I've ever met.

好吧。

Okay.

那我怎么知道区别在哪里？

Well, how the hell do I know the difference?

我为什么要认为超级智能模型的输出价值是已经比我聪明的模型的十倍？

Why would I value the super intelligent model's output 10 times more than the one that's already smarter than me?

我又怎么知道它确实好十倍？

And how do I know it's 10 times better?

我现在已经达到了另一种极限——我甚至没有足够的信息、价值观或智慧来比较这些东西。

I've now reached this kind of other limit of not even having enough information or values or intelligence to compare these things to each other.

那我为什么要用更智能的那个？

So why does it matter for me to use the more intelligent one?

实际上我认为我们遇到了一个门槛，因为模型本身永远不会有主观偏好。

I actually think we hit a threshold because the model itself is never going to have a subjective preference.

它永远不会——它不会有判断力。

It's never going to it's not gonna have judgment.

你必须是有生命的存在才能拥有判断力和价值观。

You have to be mortal to have judgment, to have values.

所以它根本不会知道或关心为什么某个模型会比另一个模型更好。

So it's simply not going to know or care why something would be like, one model would be better than another model.

而从人类的角度，从用户的角度来看，这就像已经完美无缺的计算机图形。

And from the human standpoint, from the user standpoint, it's like computer graphics that are already perfect.

好的。

Okay.

所以这个稍微好一点。

So this one's like a little bit better.

这正是我们所看到的。

And this is exactly what we saw.

就像，我们现在所处的这个阶段是超级超级增长期，大家都在竭尽全力打造更好的模型。

Like, so what we have right now is this period of like hyper hyper growth and hyper, like, doing everything you can to get a better model.

我的模型更好。

My model's better.

我的基准测试结果更好。

My the my the benchmarks are better.

一切都更好。

Everything's better.

它更大。

It's bigger.

这个更大。

This was bigger.

这个更容易获取到。

This was more available.

这个功能齐全花里胡哨，那个那个编码更强，全是些破玩意儿。

This one's got all the bells and whistles, and this one this one does code better, and all this crap.

然后就这样一遍又一遍地重复。

And it's over and over and over again.

所有东西都被逼到极限。

Everything's just push it to its limit.

我们怎么才能把它做得更大？

How can we make it bigger?

我们怎么才能把它做得更好？

How can we make it better?

我们怎么才能让它更智能？

How can we make it more intelligent?

超级智能随时可能实现。

Super intelligence is anywhere around the corner.

同样的事情发生在电子游戏和电脑图形领域，PlayStation、PS2、PS3、Xbox、任天堂，这些技术发展了十到十五年。

That same thing happened in video games and computer graphics with PlayStation, PlayStation two, PlayStation three, Xbox, Nintendo, all of these things for ten, fifteen years.

就像，那是唯一重要的事。

Like, that was the only thing.

每次有新发布会，出新游戏，出新游戏机，大家只会说'看这画质'。

Every single time there was a new conference, and there were new games, and there was a new gaming console, look at the graphics.

天呐。

Holy crap.

从2D到3D。

Two d to three d.

天啊。

Holy crap.

从3D到极致，就像，超精细的。

Three d to to, like, hyper detailed.

天啊。

Holy crap.

看看这个过场动画。

To look at the cutscenes in this.

伙计，这看起来像电影一样。

Man, this looks cinematic.

我的天啊。

Oh my god.

你甚至能看到角色脸上的汗珠。

You can see the beads of sweat on the characters.

这玩意儿火了大概十五年。

It was the thing for, like, fifteen years.

然后它就不再是视觉奇观了。

And then it stopped being a spectacle.

它停止了，只是因为已经足够好到不再重要了。

It stopped it just got good enough that it didn't really matter.

而下一代产品，你知道的，投入了10倍的GPU性能，就为了在PS4上展示更好的画面效果和过场动画之类的，但画质提升其实也就好了一点点。

And the next one that spent, you know, put 10 times as much GPU power to get, like, better graphics for the PlayStation four and showing off the cut scenes and stuff, the graphics were, like, a little bit better.

修复后的物理引擎和其他方面都更好了，但由于没人能评判，所以并未带来预期回报。

The fix physics engines were better and all this stuff, but it just didn't have the return because nobody could judge it.

没人在乎。

Nobody could care.

它看起来仍然有点像《最终幻想10》，后者确实相当酷炫。

It really still just kinda looked like like Final Fantasy 10 looked pretty pretty dope.

后来发生了什么？

And so what happened?

整个行业都变了。

The entire industry changed.

营销回报和图形竞赛的边际回报逐渐消失，人们开始思考：我们如何开发更具扩展性的游戏？

The marketing and return the marginal return on the graphics fight just kind of vanished, and now people were looking for, like, well, how can we apply the how can we make more expansive games?

我们如何在游戏中赋予更多自由？

How can we put more freedom in games?

如何创造让人不断重返的游戏环境？如何打造社交游戏体验？

How can we create an environment where people just wanna come back and play over and over and over again, where they're playing socially?