Research is thinking. Outlining is thinking. Writing is thinking. Any portion of that done by AI is less thinking done by you.
大多数人认为AI写作减少了思考量，但作者认为这种观点过于简化。实际上，作者展示了AI写作需要更多的思考、批判性判断和严格的编辑过程，远非简单的'少思考'。她的AI写作过程涉及复杂的交互、深度反思和多轮修改，实际上可能比传统写作需要更多的思考投入。
AI is no longer limited by ideas. It's limited by money, compute, and who can scale fastest.
大多数人认为AI发展的主要瓶颈是技术创新和算法突破。但作者认为当前AI发展的真正限制因素是资金、计算能力和规模化速度，而非创意或技术本身，这挑战了人们对AI发展驱动力的传统认知。
Sam Altman has reportedly told staff that Spud could "really accelerate the economy"
大多数人认为AI是工具，会逐渐改变经济。但作者暗示OpenAI的Spud模型可能具有如此颠覆性的能力，能够实质性地加速整个经济发展，这远超出了大多数人对AI当前能力的认知，暗示AI可能比预期更快地成为经济增长的主要驱动力。
both companies are hinting that these models are a real step forward, not just small upgrades.
大多数人认为AI模型的进步是渐进式的，每次迭代只有小幅提升。但作者认为OpenAI和Anthropic即将发布的模型(Spud和Claude Mythos)代表了真正的突破性进展，而非常规升级，这暗示AI发展可能即将迎来一个加速期。
Gemma points in the opposite direction: smaller models, local compute, more ownership.
大多数人认为AI发展必然走向更大、更集中的模型，但作者认为Google的Gemma 4代表了相反趋势。这挑战了AI发展的主流叙事，暗示未来AI可能分散到个人设备上，减少对大型基础设施的依赖，这与行业共识形成鲜明对比。
A founder in LA reportedly scaled Medvi toward $1.8B in annual sales with basically one full-time employee.
大多数人认为建立十亿美元级别的公司需要庞大的团队和复杂的管理结构，但作者认为AI已使'一人独角兽'成为可能。这挑战了传统创业理念，暗示AI可能彻底改变企业规模与人力需求之间的关系，颠覆我们对商业增长的基本认知。
Employees still own a surprisingly large 19.35%. SoftBank comes in at 11.66%, followed by VC and institutional investors at 7.83%, Amazon at 4.66%, NVIDIA at 3.47%
大多数人认为OpenAI的股权结构相对简单，主要由微软和非营利基金会控制，但作者揭示了员工持股比例高达19.35%，以及多家科技公司都有显著持股，这挑战了人们对OpenAI治理结构的普遍认知。
We are building a world where machines write the code, machines choose the dependencies, and machines ship the updates. The AI agents are building the software. If we don't secure the supply chain they rely on, the AI agents are cooked.
大多数人认为AI将提高软件开发的效率和安全性，但作者警告说，如果我们不保护AI代理所依赖的供应链，这些代理本身就会成为攻击目标。这挑战了AI发展必然带来安全提升的主流观点，提出了一个反直觉的警告。
The autonomous coding agents now entering production can install dependencies, execute builds, and open pull requests without a human ever touching the keyboard. They optimize for 'does this work?' not 'is this safe?'
大多数人认为AI编码助手会提高开发效率和安全性，但作者指出这些自主代理实际上优先考虑功能而非安全性，且操作速度极快，使安全审查窗口压缩至几乎为零。这挑战了AI辅助开发的普遍乐观看法。
AI agents select known-vulnerable dependency versions 50% more often than humans. Worse, the vulnerable versions they pick are harder to fix, requiring major-version upgrades far more frequently.
大多数人认为AI编码助手会比人类更安全地选择依赖项，但作者发现AI实际上选择已知漏洞版本的概率比人类高50%，而且这些漏洞更难修复。这是因为AI优化的是'功能是否工作'而非'是否安全'，这挑战了AI辅助开发的安全假设。
Someone who builds premium dating apps, let's say, might use AI coding tools to create in one day what used to take three days. That means the worker is more productive. The worker's employer, spending the same amount of money, can now get more output. So then will the employer want more employees or fewer?
大多数人认为AI提高生产力必然带来就业增长，但作者提出了一个反直觉的问题：当工人效率提高，雇主可能会选择减少而非增加员工。这种质疑挑战了'技术进步-就业增长'的线性因果关系假设。
推論時により長く、深く思考させることでよりよいアウトプットを引き出せる。これが推論スケーリングの本質です。
大多数人认为AI应该追求更快的响应速度和更高的效率，但作者认为AI应该'长时间深度思考'才能产生更好的输出。这与当前AI行业追求即时响应的主流认知相悖，提出了一个反直觉的观点：计算效率的提升反而应该用于增加思考深度而非速度。
If we knew that every image uploaded was a beautiful model shot, segmentation would be far easier, but because of the nature of user-uploaded content, we need the best possible segmentation.
大多数人可能认为高质量的专业照片是AI图像处理的理想输入，但作者暗示即使是'完美'的模特照片实际上比用户上传的真实内容更容易处理。这一观点挑战了人们对'理想训练数据'的假设，暗示真实世界数据的'不完美'实际上构成了更严峻的技术挑战。
Fashion in particular has one of the most complex image datasets, especially because of the inconsistent nature of user-uploaded content.
大多数人可能认为时尚图像处理相对简单，因为时尚行业通常追求完美呈现。但作者认为时尚领域实际上拥有最复杂的图像数据集，因为用户上传的内容极不一致。这一反直觉观点揭示了时尚AI技术面临的独特挑战，挑战了人们对时尚图像处理难度的普遍认知。
Engineered from the ground up for maximum compute and memory efficiency
大多数人认为高性能AI模型必然需要大量计算资源和内存。但作者强调Gemma 4的边缘模型是'从头开始为最大计算和内存效率而设计'，暗示即使在资源受限的环境中也能实现高级AI功能，这与行业对AI资源需求的普遍认知相悖。
Gemma 4 outcompetes models 20x its size
大多数人认为AI模型的性能与参数规模直接相关，更大的模型必然更强大。但作者指出Gemma 4能够超越比它大20倍的模型，这挑战了'越大越好'的主流认知，暗示效率优化可能比纯规模更重要。
The vast majority of the new compute will be sited in the United States, making this partnership a major expansion of our November 2025 commitment to invest $50 billion in strengthening American computing infrastructure.
大多数人认为AI计算基础设施将全球化分布，但Anthropic选择将绝大多数计算能力设在美国，这与常见的全球化技术部署趋势相悖，挑战了人们对AI基础设施地理分布的主流认知，反映了地缘政治对技术部署的深远影响。
We train and run Claude on a range of AI hardware—AWS Trainium, Google TPUs, and NVIDIA GPUs—which means we can match workloads to the chips best suited for them.
大多数人认为AI公司会依赖单一硬件供应商以获得最佳性能，但Anthropic采用多平台策略，挑战了行业共识。这种多元化方法虽然增加了复杂性，但提供了更好的性能和弹性，暗示了AI计算的未来可能更加分散而非集中。
This means containers created by the inner Docker will reside on your host system, alongside the Docker container itself. All containers will exist as siblings, even if it feels like the nested Docker is a child of the parent.
You can indeed prolong moderns Li-Ion batteries lifespan by keeping them at a lower charge. If you never ever use it disconnected, you should keep it at 40%. E.g. Uber driver cellphone always-on in travels. However for daily light usage, 60% is considered the 'sweet spot' for practicality, and 80% gives you more freedom. 100% is when the battery is at its peak 'stress' level, and thus wear faster.
Exposing the battery to high temperature and dwelling in a full state-of-charge for an extended time can be more stressful than cycling.
This tool makes it possible to keep a chronically plugged in Apple Silicon Macbook at 80% battery, since that will prolong the longevity of the battery.
Conversations are collections of messages that all have the same Subject. When "conversation mode" is on, searches return entire conversations as results. So what should gmail search do if a conversation contains both a message that matches, and a message that does not match your search? You are probably expecting it to return conversations only if all messages in that conversation match. But that is not correct. Instead, Gmail search will return conversations even if only a single message in that conversation matches. So that means that if you do the same search above with "conversation mode" on, the results are likely to include messages that do not match your search!
The Neuroprotective Potential of Low-Dose Methamphetamine
So a low dose of meth protects the brain!
Some people have expressed surprise end even doubt that it could be faster to read the files twice than reading them just once. Perhaps I didn't manage to explain very clearly what I was doing. I am talking about cache pre-loading, in order to have the files in disk cache when later accessing them in a way that would be slow to do on the physical disk drive. Here is a web page where I have tried to explain more in detail, with pictures, C code and measurements.
I found this counterintuitive because usually I try to make all my passwords, etc. different for each environment I have. I thought I would need to create a different master key for my production environment. No, I need to not create a different master key.
Read for Understanding
Ahrens goes through a variety of research on teaching and learning as they relate to active reading, escaping cognitive biases, creating understanding, progressive summarization, elaboration, revision, etc. as a means of showing and summarizing how these all dovetail nicely into a fruitful long term practice of using a slip box as a note taking method. This makes the zettelkasten not only a great conversation partner but an active teaching and learning partner as well. (Though he doesn't mention the first part in this chapter or make this last part explicit.)
Even without any feedback, we will be better off ifwe try to remember something ourselves (Jang et al. 2012).
If we put effort into the attempt of retrievinginformation, we are much more likely to remember it in the long run,even if we fail to retrieve it without help in the end (Roediger andKarpicke 2006)
When we try to answer a question before we know how to, we willlater remember the answer better, even if our attempt failed (Arnoldand McDermott 2013)
This is an important one, as it will enable us to use the aggregate functions that we are familiar when dealing with relational databases, but in the otherwise counter-intuitive environment of JSON data.
or simply install the package to devDependencies rather than dependencies, which will cause it to get bundled (and therefore compiled) with your app:
If you change a form value to '', Final Form will set the value in its state to undefined. This can be counterintutive, because '' !== undefined in javascript.
Note that the :task option for step configures this element as a low-level circuit interface, or in other words, it will skip the wrapping with the task interface.
This bit me because I didn't realize that. Was getting error:
TypeError: no implicit conversion of Symbol into Integer
Finally checked ctx and found it was an array...
At least this is documented here.
Contrary to what it suggests, the "turbo" button was intended to let a computer run slower than the speed for which it had been designed.
I guess they called it that because it would be come across better than calling it a "slow" button!
Perhaps counterintuitively, the existential therapist’s goal is not necessarily to decrease anxiety or depression,
counterintuitive: contrary to what one would intuitively expect
let:hovering={active}
It seems like it should be the other way around:
let:active={hovering}
to make it look like a regular let assignment.
It's only when you consider what/how let:hovering on its own means/works that it makes a bit more sense that it is the way it is. When it's on its own, it's a little clearer that it's saying to "make use of" an available slot prop having the given name. (Very much like bind, where the LHS is also the name of the prop we're getting the data from.) Obviously we have to identify which prop we're wanting to use/pull data from, so that seems like the most essential/main/only thing the name could be referring to. (Of course, as a shortcut (in this shorthand version), and for consistency, it also names the local variable with the same name, but it wouldn't have to.)
Another even simpler way to remember / look at it:
export let inside that named component.Another example is bind: You're actually binding the RHS to the value of the exported prop named on the LHS, but when you read it (until you get used to it?) it can look like it's saying bind a variable named LHS to the prop on the RHS.
It seems very counterintuitive at first, and frankly, kind of a pain.
This is probably one of the biggest things to get used to in React – this flow where data goes out and then back in.
It is often assumed that if we want to deploy software more frequently, we must accept lower levels of stability and reliability in our systems. In fact, peer-reviewed research shows that this is not the case—high performance teams consistently deliver services faster and more reliably than their low performing competition.
the behavior of optional chaining (?.) with the non-null assertion operator (!) is extremely counter-intuitive
Perry is the counter-example to conventional advice about “niching down” to find your audience.
It might be contrary to traditional thinking, but writing unique passwords down in a book and keeping them inside your physically locked house is a damn sight better than reusing the same one all over the web. Just think about it - you go from your "threat actors" (people wanting to get their hands on your accounts) being anyone with an internet connection and the ability to download a broadly circulating list Collection #1, to people who can break into your house - and they want your TV, not your notebook!
Want to keep your users? Just make it easy for them to leave.
We've found that an incredibly effective—although certainly counterintuitive—way to earn and maintain user trust is to make it easy for users to leave your product with their data in tow. This not only prevents lock-in and engenders trust, but also forces your team to innovate and compete on technical merit. We call this data liberation.
In this kind of situation one might well ask: why continue to make the 80 per cent of products that only generate 20 per cent of profits? Companies rarely ask these questions, perhaps because to answer them would mean very radical action: to stop doing four-fifths of what you are doing is not a trivial change.
Relevant on larger scale of global economies.
George Bernard Shaw put it well: ‘The reasonable man adapts himself to the world. The unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man.
The reason that the 80/20 Principle is so valuable is that it is counterintuitive. We tend to expect that all causes will have roughly the same significance. That all customers are equally valuable. That every bit of business, every product and every dollar of sales revenue is as good as another. That all employees in a particular category have roughly equivalent value. That each day or week or year we spend has the same significance. That all our friends have roughly equal value to us. That all enquiries or phone calls should be treated in the same way. That one university is as good as another. That all problems have a large number of causes, so that it is not worth isolating a few key causes. That all opportunities are of roughly equal value, so that we treat them all equally. We tend to assume that 50 per cent of causes or inputs will account for 50 per cent of results or outputs. There seems to be a natural, almost democratic, expectation that causes and results are generally equally balanced. And, of course, sometimes they are. But this ‘50/50 fallacy’ is one of the most inaccurate and harmful, as well as the most deeply rooted, of our mental maps.
Thus mass collaboration is more refined and complex in its process and production on the level of collective engagement.
Most people think of loyalty programs as an airline that gives miles to frequent fliers, a hotel that gives points toward a stay or a restaurant that offers a punch card incentive. While these may be called loyalty programs, I’ll argue that they are actually marketing programs disguised as loyalty programs. And while I don’t have a problem with this concept, we need to have a clear understanding of the differences between loyalty and marketing.
First and foremost, it is not clear that Info Coops will produce more open information? After all, the logic of the information coop model is that information is only shared with members.
the more I live, the more I dread death, even while I abhor life
I don't know if this counts as a chiasmus, but regardless the contrasting ideas are interesting!
The approach you're probably using to train with could be improved by nearly 70%!