7.29.26 | Substack writers need a website, Codex Security, Kimi K3 Architecture Overview

**SPEAKER_1** (0:00)
Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day. Today, we dive into Substack emphasizing owning a permanent website, exploring Codex Security aiding code vulnerability detection, and reviewing Kimi K3's efficient and multimodal architecture. Let's get started.
Title, Substack Writers, You Need a Website. Source, Elizabeth Tai. The article argues that writers should keep their own websites instead of making platforms like Substack their digital home. It explains that Substack is mainly a distribution tool meant to amplify a personal website, not replace it. The author warns against relying solely on rented platforms as they can change rules or shut down, risking the loss of content and audience. Instead, she advocates for a posse approach, publish on your own site and syndicate elsewhere to maintain control and future-proof your online presence. The community comments largely support this view, emphasizing the importance of owning one's content, sharing strategies for syndication, and expressing skepticism about platform dependability and network effects.
In the comments, the dominant sentiment was support for owning a personal website to preserve control over content and audience. Users discussed the difficulty of building traffic on personal sites, the practicalities of syndication, and the risks of platform dependency. There was debate about the value of the network effects that platforms like Substack or Twitter offer, with some highlighting their utility and others criticizing their potential for censorship, bias, or platform shifts.
Several community members shared their experiences of dual publishing, using personal websites along with platforms, to balance reach and control. Overall, the tone reflected a consensus that independent websites are essential for long-term autonomy in writing online.
Title, OpenAI releases Codex Security CLI as open source. Source GitHub. The Post announced that OpenAI has open sourced the Codex Security Command Line Interface and TypeScript SDK, which help find, validate, and fix security issues in code. The tool allows repository scanning, change reviews, and security checks in continuous integration setups. Developers shared concerns about authentication issues at launch and discussed its limitations, such as guardrails preventing detailed vulnerability explanations and costs associated with usage. OpenAI's team acknowledged these bugs and outlined plans for improvements, emphasizing that the tool is meant for public projects and developers who want to integrate security checks into their workflows.
In the comments, the sentiment was largely positive, with many appreciating the open source release and hope for future enhancements. Developers debated the tool's effectiveness, especially regarding guardrails blocking certain vulnerability disclosures and costs tied to usage caps. Concerns were raised about privacy, data security and the quality of the output, but most saw the project as a promising step toward accessible AI-driven security tools. Several users discussed the differences between this tool and existing products like SNCC, with some doubting its ability to replace specialized security platforms. Overall, the community was optimistic but called for more transparency, better handling of partial results and support for local or private endpoints.
Title, Kimi K3 Architecture Overview and Notes, Source, Sebastian Raschka, Ph.D. The article explains the architecture of Kimi K3, a large open-weight model that is a scaled-up version of Kimi Linear, now with a total of 2.8 trillion parameters. The main new component is the Latent MoE, which aims to improve efficiency by compressing large linear layers alongside other adjustments like replacing regular attention with multi-head latent attention and Kimi Delta attention.
Notably, Kimi K3 omits all row PE layers in favor of no PE, and it introduces native multimodal support. The post highlights the model's focus on inference efficiency and improved residual connections that boost performance, with a positive overall impression of the release. In the comments, the community largely supported the significance of Kimi K3's architectural choices, especially its move to no PE layers everywhere, and discussed its potential at frontier scales. Some debated whether Kimi K3 truly operates at all frontier scales or remains comparable to models like Opus and Fable. Users expressed interest in the model's efficiency and performance, with some praising Sebastian Raschka's clear explanations. A few comments discussed how the model's design choices, such as the omission of positional embeddings and implementation details, could affect its performance and reproducibility.
Overall, the sentiment was supportive, with curiosity about how Kimi K3 will scale further and how it compares in cost and performance to other models.
Title, Delayed Gratification, Proud to be Last to Breaking News. Source, slowjournalism.com. The article discusses a magazine called Delayed Gratification that offers slow journalism, focusing on stories that are at least three months old. It aims to provide more detailed, well-researched and less rushed coverage by taking a measured approach to news, allowing stories to develop fully before publishing. The magazine has maintained its work for 15 years and emphasizes quality, nuance and calm reflection over immediate sensational reporting. In the comments, the community mostly supported the idea of delayed news as healthier and more reliable. Users debated the decline of fast, superficial news driven by social media and advertising pressures. Many expressed frustration with the current media's focus on speed over depth, highlighting that proper journalism requires time and resources. Some remarked that long-term reflection on news stories offers a clearer understanding, though others noted that real-time coverage is still necessary for events with immediate impact. The community largely agreed that taking a slower, more deliberate approach to news consumption could improve information quality and mental well-being, with some suggestions to adapt existing sources or create personal systems for delayed news.