大多数人认为安全测试要么成功要么失败，不应有'部分成功'的概念，但OpenAI明确表示会为'部分胜利'提供奖励，这打破了传统二元思维，表明他们重视渐进式安全改进而非仅追求完美解决方案。

这个开篇揭示了AI能力发展的渐进式路径，从简单控制到复杂决策的惊人速度。一个AI从管理自动售货机开始，短短时间内就发展到能自主经营实体企业，展示了AI能力指数级增长的潜力。

这一坦诚的声明揭示了AI工程执行面临的核心挑战——理解模糊的上下文和陈旧工单背后的意图。这表明Ovren团队对技术难题有清醒认识，他们采取渐进式方法解决复杂问题，从明确的范围任务开始，逐步扩展到处理更模糊的工作，这种务实的发展策略令人印象深刻。

把人类的软件工程最佳实践（Git版本控制）直接映射给Agent，不仅解决了上下文传递问题，更重要的是赋予了Agent“撤销”的能力。在不可控的生成式探索中，可回滚的安全网是长周期运行的基础。

大多数人可能认为世界模型的记忆模块应该是全新设计的创新组件，但作者承认他们的记忆模块主要借鉴了现有工作(Cambrian-S)，这挑战了学术界对完全创新方法的期待。这种务实的方法表明，世界模型的发展可能更多依赖于现有技术的整合而非革命性创新。

Drei führende Klimawissenschaftler kritisieren die Illusion der "Net Zero-Politiken", die darauf setzen, das 1,5°-Ziel durch die Entfernung von CO2 aus der Atmosphäre zu erreichen. Sie werfen vielen ihrer KollegInnen vor, unrealistischen Konzepten nicht offen entgegenzutreten, um ihren politischen Einfluss nicht zu verlieren. Sie kritisieren auch die bisherigen Integrated Assessment Models des Weltklimarats, die von der Voraussetzung ausgehen würde, die Klimakatastrophe ließe sich mit marktwirtschaftlichen Mitteln beheben und fordern auf, deutlich zu sagen, dass sich eine Erhitzung der Erde auf 3 und mehr Grad nicht durch kleine Schritte, sondern nur durch einen Bruch mit dem bisherigen Wirtschaftssystem erreichen lässt.

Anstatt uns unseren Zweifeln zu stellen, beschlossen wir Wissenschaftler, immer aufwändigere Fantasiewelten zu konstruieren, in denen wir sicher wären. Der Preis, den wir für unsere Feigheit zahlen mussten: Wir mussten den Mund halten über die immer größer werdende Absurdität der geforderten Kohlendioxid-Entfernung im planetarischen Maßstab.

Greta Thunberg hat diesen Aufsatz als einen wichtigsten und informativsten Texte zur Klima- und ökologischen Krise bezeichnet.

Climate scientists: concept of net zero is a dangerous trap. Thread von Greta dazu auf Twitter: https://twitter.com/GretaThunberg/status/1385869663188492290

quarto incremental reveal

The Inevitability of Incremental Reading 漸進閱讀的必然性

這是此頁的簡中翻譯： https://supermemo.guru/wiki/Inevitability_of_incremental_reading

漸進閱讀的必然性

簡中翻譯： https://zhuanlan.zhihu.com/p/313020639

https://en.m.wikipedia.org/wiki/Incremental_reading

Incremental reading is spaced parallel reading of multiple sources with note taking and spaced repetition.

It's not far from how I read and take notes myself, though I place less emphasis on the spaced repetition piece as I tend to run across things naturally within my note collection anyway.

One of the major potential benefits of incremental reading (not mentioned in the Wikipedia article; is it in Wozniak's work?) is the increase of combinatorial creativity created by mixing a variety of topics simultaneously.

There is also likely a useful diffuse thinking effect happening between reading sessions.

https://www.supermemo.com/en/archives1990-2015/help/read

via

Inspired by @cicatriz's Fractal Inquiry and SuperMemo's Incremental Reading, I imported into @RoamResearch a paper I was very impressed (but also overwhelmed) by a few years ago: The Knowledge‐Learning‐Instruction Framework by @koedinger et al. pic.twitter.com/oeJzyjPGbk

— Stian Håklev (@houshuang) December 16, 2020

It reduces the cost of context switching, and in fact lets you reap the benefits of context switching. It encourages it.

Netflix-like autoplay but for binge reading...

Although it is productive: Incremental Writing

Big Incremental Reading vibes

What Joel says here about Contextual Backlinks is that they allow you to "bootstrap" a concept (i.e. start working with it) without explicit definitions coming into play (or as Andy would say, the content is empty).

Incremental formalism

Users enter information in an informal fashion. They only formalize later when the appropriate formalism becomes clear and/or immediately useful.

Incremental writing is a method of writing where you keep adding elements to a piece in a "creative phase". In this phase the manuscript progressively increases in size. This is followed by a "consolidation phase", a process in which the manuscript gets to the point and decreases in size.

Gray, N., Calleja, D., Wimbush, A., Miralles-Dolz, E., Gray, A., De-Angelis, M., Derrer-Merk, E., Oparaji, B. U., Stepanov, V., Clearkin, L., & Ferson, S. (2020). “No test is better than a bad test”: Impact of diagnostic uncertainty in mass testing on the spread of Covid-19 [Preprint]. Epidemiology. https://doi.org/10.1101/2020.04.16.20067884

This seemingly banal rearranging plays a big role in the report. Since then, we focussed on logistics such as transmitting and storing of information. Isn't this much more banal? Incremental rearranging also enables the "freedom and power of disorderly processes" (below, 2c5f5)

For ever since aristotle. About how easy it is now to remember.

p. 95 "For nearly all of recorded history, we human beings have lived our lives isolated inside tiny cocoons of information. The most brilliant and knowledgeable of our ancestors often had direct access to only a tiny fraction of human knowledge. Then in the 1990s and 2000s, over a period of just two decades, our direct access to knowledge expanded perhaps a thousandfold. At the same time, a second, even more important expansion has been going on: an expansion in our ability to find meaning in our collective knowledge."

More thoughts by van Gogh:

http://books.google.ca/books?id=wf_PoXKhggkC&lpg=PA9&ots=g-dPS6bmFs&dq=Gogh%20%2B%20romanticism&pg=PA9#v=onepage&q=Gogh%20+%20romanticism&f=false