这里存在未经证实的假设，作者声称GPT对LGBT内容更宽松，但没有提供任何证据支持这一说法。这种断言可能基于有限的个人观察或选择性案例。改进方法应该是提供具体的测试数据或研究结果来支持这一假设，或者明确指出这只是基于个人经验的观察而非普遍事实。

文章声称Chrome没有4GB模型文件也能正常运行，但没有提供证据支持这一断言。虽然Chrome可能在某些功能上不依赖该模型，但完全移除可能影响性能或某些功能。需要更详细的分析来说明模型与Chrome核心功能之间的关系，而不是简单地假设它是可选的。

文章做出了一个具体的环境影响断言，但没有提供详细的计算过程或数据来源。虽然引用了Pärssinen等人的研究，但将研究结果应用到Chrome的具体规模上时缺乏透明度。改进方法应包括完整展示计算公式、所有假设条件以及数据来源，以便读者能够验证这些数字的准确性。

作者声称应用程序不应静默修改另一个应用程序是一个'明显'的原则，但并没有提供支持这一原则的行业标准、法律先例或广泛共识。这是一个未经证实的假设，可能反映了作者的个人观点而非行业共识。改进方法应提供支持这一原则的权威来源，如行业指南、法律先例或广泛认可的最佳实践。

作者将Anthropic自己测量的prompt注入成功率(11.2%有防御措施后)直接应用到这个桥接功能上，但没有提供证据表明这个特定功能具有相同的漏洞率。这是一个未经证实的假设，将一般性安全数据应用到特定功能上。改进方法应包括提供针对这个特定桥接功能的实际安全测试数据，或者明确说明这是基于Anthropic一般性安全数据的推测。

作者对Anthropic所说的“仅对2%的新用户进行小规模测试”的说法表示怀疑，这表明可能存在更大的影响范围。

This statement suggests a reliance on user acceptance of a potentially invasive technology, which may be an overestimate of public willingness or a speculative assumption.

这一描述揭示了早期数据代理的简化假设：将问题简化为自然语言到SQL的转换。这挑战了仅通过改进模型性能就能解决所有数据推理问题的乐观预期，强调了业务语义理解的重要性。

for - quote - who you are beyond your headset - Donald Hoffman - If you want to understand the truth of who you are beyond just this headset description of you - then you have to - lay aside all concepts period and - just know yourself by being yourself, - not by putting a concept between you and yourself. - adjacency - headset - perspectival knowing - Donald Hoffman - unquestioned assumption of other perspectives - imputation - external observable proxy - to private, inner world - As I read Hoffman's use of the word "headset", it brought up some associations with the idea of "perspectival knowing" - There is the perspectival knowing of a species, - but also of the individual of a species - For humans, perspectival knowing must be contextualized within an imputation: - that other perspectives exist - in other words, that other private worlds exist - and ultimately, this is a widely accepted imputation of an inner private world - based upon public, external observable behavioral proxies - This imputation of the other is a fundamental imputation and assumption of the human condition which we all take for granted, - but because it is so foundational, never question

An assumption, like many others (such as the bathing suit situation) about Oliver's identity that is quickly refuted, because identities never make sense. A person as a whole cannot be summarized in rules or statements or if.. then.. conditions.

\((i, j)\in A\iff (j, i)\in A\). We allow for arcs with zero capacity. For all single direction arc, the opposite direction arc on the graph is having a capacity of zero.

I think this is essentially a residual graph on the zero flow, which is trivially feasible. Residual graph on the network is covered in the later part of the chapter (pg177).

No parallel arcs.

Doesn't contain a direction path with infinite capacity from source to node. (Makes sure that the problem is not unbounded)

All capacities are nonnegative integers.

The Network is Directed

Assumptions required for labeling algorithms, and non-directed edges can be reduced to parallel double edges.

Koerth, M. (2021, November 3). The Science You Need To Make Your COVID-19 Decisions. FiveThirtyEight. https://projects.fivethirtyeight.com/covid-19-updates/

"unchallenged truths" is not really accurate. More like unchallenged assumption.

Adam Kucharski [@adamjkucharski]. (2021, September 8). Some tips on interpreting models (from an @SMC_London talk I gave a few months ago): Https://t.co/3NlRN6q6gb [Tweet]. Twitter. https://twitter.com/adamjkucharski/status/1435650792082575360

Editor, L. J., Health. (2021, August 22). Hundreds of doctors sign open letter to PM: Need debate on “flawed covid guesses.” Express.Co.Uk. https://www.express.co.uk/news/uk/1480245/coronavirus-news-doctors-sign-letter-boris-johnson

‘Trust the science’ is the mantra of the Covid crisis – but what about human fallibility? | Margaret Simons | The Guardian. (n.d.). Retrieved July 27, 2021, from https://www.theguardian.com/commentisfree/2021/jul/24/trust-the-science-is-the-mantra-of-the-covid-crisis-but-what-about-human-fallibility?CMP=Share_iOSApp_Other

ReconfigBehSci on Twitter: ‘I just had cause to revisit the Friston modelling paper from Sept: Https://t.co/QOTC8fXV0n 1/n’ / Twitter. (n.d.). Retrieved 26 February 2021, from https://twitter.com/SciBeh/status/1336277391233208320

Lee Kennedy-Shaffer. (2021, May 6). .@rebeccajk13 and @mlipsitch showed how to use that to estimate efficacy against prevalent infection (https://t.co/LeXxqcumGS). [Tweet]. @LeeKShaffer. https://twitter.com/LeeKShaffer/status/1390322501817880581

Adam Kucharski. (2021, April 14). If populations are highly vaccinated, we’d expect a higher proportion of future cases to have been previously vaccinated (because by definition, there aren’t as many non-vaccinated people around to be infected). But what sort of numbers should we expect? A short thread... 1/ [Tweet]. @AdamJKucharski. https://twitter.com/AdamJKucharski/status/1382242089673617417

Golding, S. E. (2020). Coronavirus and other pathogens: Reflecting on the relationship between health psychology and infectious disease [Preprint]. PsyArXiv. https://doi.org/10.31234/osf.io/8r6kf

Is the author assuming that most people regard Google Searches as anything more than a result? Gospel? How many people have even a rudimentary understanding of page rank? Whose duty is it make sure they do?

That’s something we’ve been doing, yes.

Among the most useful summaries I have found for Linked Data, generally, and in relationship to libraries, specifically. After first reading it, got to hear of the acronym LODLAM: “Linked Open Data for Libraries, Archives, and Museums”. Been finding uses for this tag, in no small part because it gets people to think about the connections between diverse knowledge-focused institutions, places where knowledge is constructed. Somewhat surprised academia, universities, colleges, institutes, or educational organisations like schools aren’t explicitly tied to those others. In fact, it’s quite remarkable that education tends to drive much development in #OpenData, as opposed to municipal or federal governments, for instance. But it’s still very interesting to think about Libraries and Museums as moving from a focus on (a Web of) documents to a focus on (a Web of) data.

The “Open World Assumption” is central to this post and to the actual shift in paradigm when it comes to moving from documents to data. People/institutions have an alleged interest in protecting the way their assets are described. Even libraries. The Open World Assumption makes it sound quite chaotic, to some ears. And claims that machine learning will solve everything tend not to help the unconvinced too much. Something to note is that this ability to say something about a third party’s resource connects really well with Web annotations (which do more than “add metadata” to those resources) and with the fact that no-cost access to some item of content isn’t the end of the openness.

Enthymeme? Is this idea an assumption held by the public?