1. Things I won’t work with: the higher states of bromine (2019)
Total comment counts : 25
Summary
error
Top 1 Comment Summary
The article is part of Derek Lowe’s “Things I won’t work with” series on his blog. It discusses the chemical compound Dioxygen Difluoride (FOOF) and recommends another post about it from 2010. The article is written in an accessible and humorous way, making it enjoyable for both chemistry and non-chemistry enthusiasts.
Top 2 Comment Summary
The article is a recommendation for a series of humorous science writing by Derek Lowe. The writer praises the articles as some of the funniest they have ever read in the genre. The series, titled “Things I Won’t Work With”, consists of 33 articles written over two decades. The writer expresses a desire for more entries in the series, although Derek Lowe only occasionally posts new articles. The article provides a link to a starting point for the series and an index of all the “Things I Won’t Work With” posts.
2. Some notes on Rust, mutable aliasing and formal verification
Total comment counts : 13
Summary
The article asks the reader to complete a CAPTCHA to verify their requests and mentions Dreamwidth Studios’ copyright.
Top 1 Comment Summary
The article discusses the usefulness of the Rust programming language for applying formal methods in program verification. It mentions that Rust’s rules make it easier to formalize cases that are typically difficult. However, there are still challenges in deadlock analysis, both in the thread and borrow sense. The author expresses the desire for static deadlock analysis in Rust, which would allow for safe back-pointers and eliminate the need for most reference counts. The article also comments on the limitations of theorem provers and suggests that machine learning could assist in guiding the theorem proving process.
Top 2 Comment Summary
The article discusses a quote from a paper by Hoare from around 1973 and how it was being skewed towards a Rust-centric view. The author expressed their disagreement with this narrowing of the scope of Hoare’s critique. However, they acknowledge that the subsequent discussion in the blog post by Grayson addressed the technical issues raised in the original article.
3. VCs aren’t your friends
Total comment counts : 58
Summary
The article discusses a recent incident involving VC Jason Lemkin, who passed on a startup because the pitch deck was dated from two months ago. The author argues that Lemkin did nothing wrong and highlights the misconceptions that founders often have about VC relationships. The author explains that VCs receive numerous pitch decks and rely on small details and signals to make quick elimination decisions. Founders, on the other hand, may expect more support and feedback from VCs, but they need to understand that VCs are primarily focused on their own financial interests. The article concludes that while some founders may have been offended by Lemkin’s decision, it was actually helpful for others to understand the ruthlessness and bias involved in the VC selection process.
Top 1 Comment Summary
The article suggests that venture capitalists (VCs) and Ivy League schools share similar filters in selecting investments and students, respectively. These filters are not based on technical brilliance or innovation but primarily on being smart enough, following instructions, and fitting into institutional structures. While this may be appropriate for companies seeking venture capital, it potentially excludes unconventional and brilliant individuals who do not conform to these filters. The author questions whether iconic figures like Steve Jobs or Nikola Tesla would have passed these VC/Ivy League tests.
Top 2 Comment Summary
The article argues that most business founders do not actually need venture capital (VC) funding and can be worse off by taking it. The author believes that the idea of not receiving funding because of an outdated pitch deck is unrealistic and that the majority of businesses can be built without VC money. Instead, the author suggests building a solid, profitable business that is ready to scale up before considering VC funding.
4. SSD death, tricky read-only filesystems, and systemd magic?
Total comment counts : 14
Summary
error
Top 1 Comment Summary
The article advises against interacting with a filesystem on questionable media before making a copy of it. Mounting or repairing the filesystem can induce more writes than expected, even in read-only mode. It suggests using tools like “dd,” “ddrescue,” or redirection in combination with pv/cat/shell to evacuate the data first. This approach offers infinite tries and increases the chances of preserving the data on a new, reliable storage medium. The article emphasizes the importance of thinking in terms of files rather than devices and warns that by directly interacting with the filesystem, there is a risk of irreversibly breaking it. The author criticizes those who overlook these precautions and highlights the potential damage caused by inaction. They urge readers to exercise caution and prioritize data preservation. The article concludes by emphasizing the significance of mechanical sympathy when handling data.
Top 2 Comment Summary
The author shares an anecdote about running FreeBSD on a failing USB stick for about a year. The stick had multiple copies of the files and ZFS made attempts to fix any issues during runtime. The author regularly ran a scrub to check the system’s health, and although it reported problems, the system continued to function. This experience suggests that flash storage can endure a relatively high failure rate for longer than anticipated.
5. Llama 3 implemented in pure NumPy
Total comment counts : 14
Summary
The article provides an introduction to the book “자바 알고리즘 인터뷰 with 코틀린” which is a resource for developers preparing for coding tests and algorithm interviews. It mentions that the book is an upgraded version of “파이썬 알고리즘 인터뷰” and includes Java and Kotlin code. The article also briefly discusses the Llama 3 model, its structure and parameters, and the implementation of various techniques such as RoPE embedding and RMSNorm normalization. It mentions the use of GQA (Grouped-query) in Llama 2 for better performance.
Top 1 Comment Summary
The article mentions that the original implementation by Meta is only 300 lines of readable code. The link provided leads to the model.py file on GitHub.
Top 2 Comment Summary
The article is about a trainable Llama-like transformer that is implemented using only the numpy library and contains around 600 lines of code. The code implementation can be found on the GitHub repository mentioned in the article.
6. Sprint, T-Mobile Merger Killed Wireless Price Competition in U.S.
Total comment counts : 28
Summary
The article discusses the negative consequences of the T-Mobile and Sprint merger, which activists and critics had warned about before the deal was approved by the Trump administration. The article claims that the merger resulted in layoffs, less competition, higher prices, and a lower quality product in the telecom sector. It also states that a new report by a research firm found that the merger stopped meaningful price competition in the U.S. wireless market, making it one of the most expensive in the world. The article suggests that the merger was intended to benefit investors and increase profits through nickel-and-diming consumers. The Trump administration is criticized for not reading deal impact reports and working with the companies to ensure approval. The article also mentions the involvement of former T-Mobile CEO John Legere and the influence of spending at Trump hotel properties in sealing the deal. It criticizes the Trump administration for its lack of antitrust enforcement and highlights the previous decision by the Obama administration to block AT&T’s attempted acquisition of T-Mobile, which protected competition and consumer welfare. The article concludes by questioning the accountability of T-Mobile executives and the actions of the Trump administration.
Top 1 Comment Summary
The author expresses a dislike for mergers but believes that Sprint’s situation was so dire that a merger was necessary. They suggest that if the merger had not occurred, Sprint may have filed for bankruptcy and Verizon and AT&T would have benefited. The author emphasizes the severity of Sprint’s situation and considers the merger as a necessary evil.
Top 2 Comment Summary
The article discusses a government condition imposed on the merger of T-Mobile. The condition required T-Mobile to offer low-cost prepaid plans without the use of MVNO (Mobile Virtual Network Operator). T-Mobile introduced a program called T-Mobile Connect to meet this requirement. The program offers monthly plans starting at $15 for unlimited talk/text and 5G/LTE data. Additional data options are available at different price points. The article also mentions that T-Mobile’s pSIM Wi-Fi calling/texting feature will still work outside the USA with cellular/eSIM data from a second line. Additionally, the article mentions a company called Eskimo that offers data eSIMs for a price of about $4/GB.
7. Localization in .NET console and desktop apps
Total comment counts : 12
Summary
The author has found a good solution for localizing .NET applications by using the .po/.pot format for translations. They discovered a library called POTools that allows for extracting translatable strings from .resx files. They then used Loco as a translation tool and created a separate resource file for texts that should never be translated. They also used IStringLocalizer and the Karambolo.PO library to implement a POStringLocalizer. This solution allows for the use of the gettext universe of tools and users while still maintaining compatibility with .NET code and libraries.
Top 1 Comment Summary
The author wonders why translations often require an extra format or framework. They suggest using compile time code in the language used, based on a common format such as gettext with macros. The author believes that a list or map of key-value pairs, with a unique translation identifier as the key and the translation as the value, is the best, easiest, and fastest solution. They also mention the need for unique identifiers for plural forms, as automatic pluralization has failed for their use cases in Russian.
Top 2 Comment Summary
The article discusses a localization framework that was developed during the .NET 2.0 days and is still being used. The framework involves version-numbered translations stored in an XML file with a graphical user interface (GUI) to manage missing and outdated translations. The article provides links to the framework, sample translations, a guide with screenshots, and a validation example for the XML file. It also mentions that while XML is used for validation purposes, JSON is not supported without a dependency under .NET Framework.
8. F* – A Proof-Oriented Programming Language
Total comment counts : 11
Summary
The article discusses the programming language F*, which is a proof-oriented language that supports both functional and effectful programming. F* combines dependent types with proof automation and interactive theorem proving. It can be compiled to OCaml and can also be extracted to F#, C, Wasm, or assembly. F* is open source and actively developed by Microsoft Research, Inria, and the community. The language is distributed under the Apache 2.0 license and has binaries available for Windows, Linux, and Mac OS X. It can be installed from various sources and there is an online book available for learning. F* is frequently taught at seasonal schools and there are lectures and course materials available. The article also provides information on various forums, seminars, and mailing lists where users can interact and get support for F*.
Top 1 Comment Summary
The author expressed an interest in using F* for production work, but never received approval. They found F* appealing because it satisfied their desire for a language compatible with F# that also provided IDE support. While they have only explored F* through toy examples on the website, they consider it to be a more approachable system compared to Coq or Agda, and appreciate its use of dependent types. Despite their PhD work being in Isabelle, they believe that dependent types have more direct application in “real code” and are interested in utilizing F* for its .NET integration.
Top 2 Comment Summary
The author expresses a desire for a programming language that allows users to prove and confirm that their code follows a formal specification. While interested in Lean 4, they discover that it is still a research project and not yet suitable for production use, as it expects further breaking changes. They hope to one day write proofs instead of tests in certain situations.
9. Swift sucks at web serving or does it?
Total comment counts : 15
Summary
The article discusses a web server comparison conducted by Axel Roest, which initially had some incorrect conclusions. The focus of the comparison was on efficiency rather than brute performance, with a consideration for factors such as throughput, latency, RAM usage, and power consumption. The load used for the benchmark was Fibonacci sequence calculation. The article emphasizes the importance of whether a benchmark is useful rather than how well it represents the real world. The hardware used for the benchmark varied between old and newer models. The article also highlights the need to question and understand benchmark results to improve their quality and usefulness. It points out a logical error in the benchmark tool and suggests starting with basic analysis before jumping to complex conclusions. The involvement of relevant experts on the Swift Forums initially led to assumptions about the problem being with Vapor or its configuration, but it was later discovered that those were not the root issues.
Top 1 Comment Summary
The article discusses the use of the Fibonacci calculation as a benchmark for web servers and the criticisms it received. Some argue that the CPU-heavy nature of the calculation is not representative of typical web server tasks, while others claim that the benchmark is unfair to Swift because it lacks optimization for the “BigInt” implementation used in other languages. The author suggests that instead of defending the benchmark, a more relevant and fair benchmark should be created to evaluate Swift’s suitability for web serving. They also question the importance of BigInts in web serving, as they have rarely been used in their experience. The article concludes by recommending the development of a better benchmark rather than making excuses for the current one.
Top 2 Comment Summary
The author suggests that if more investment had been made in Linux support and a good standard library, Swift could have been a major player in web development 9-10 years ago. However, they believe that currently, Go is seen as more of an alternative to web development, with Rust being used in extreme cases. The author points out that they wanted to like Swift, but felt that there was not enough investment in the language outside of the Apple ecosystem. They also question why Swift has a package manager if Xcode does not use it.
10. Earth rotation limits in-body image stabilization to 6.3 stops (2020)
Total comment counts : 22
Summary
error
Top 1 Comment Summary
The article discusses the correct measurements for a day and the rotation of the Earth. It mentions that a day is not exactly 86,000 seconds but 86,164.091 seconds, and the rotation speed is not 7.27x10^-5 radians per second but 7.29x10^-5 radians per second. The article explains that it takes 24 hours for the sun to return to the same spot in the sky due to the Earth’s rotation and revolution around the sun. It also mentions that a sidereal day for distant stars to return to the same spot is 23 hours, 56 minutes, and 4.091 seconds. The author relates this information to a photography mishap at their daughter’s graduation.
Top 2 Comment Summary
This article discusses the use of inertial navigation systems (INS) in aircraft and how they can function without relying on GPS. INS use stabilised platforms and complex calculations to determine aircraft position. While it is helpful to have a rough indication of the current latitude at startup, it is also possible to calculate it using gyro outputs, but this takes longer. Modern sensors, such as solid-state laser gyroscopes, have made INS smaller and more compact, to the point where they could be used in cameras. However, the cost may outweigh the benefits, as modern cameras already have significant stabilization capabilities.