These are some things I’ve wandered across on the web this week.

🔖 tapes.01

Minimalistic, as simple as it can be, the fewer pages and tabs it has the better - that was our focus when we were designing the interface. Tapes interface is divided into two pages, first page contains only essential controls - volume, sample start, macro controls - just so you could start shaping the sound right away without worrying about the details. In the end it’s the first thing that you’ll see after loading most of the presets, and it’s a nice way to quickly find a sound you’re looking for or to shape further.

🔖 CAMP

CAMP runs five-day arts, music, writing and arts-activist sessions. These are no ordinary workshops - they are intense, artistic catalysts run by internationally acclaimed practitioners; creative flashpoints designed to change the lives of everyone involved. The workshops combine work in our well equipped facilities with projects carried out in the mountains - check out the workshops for full details.

🔖 Meta is earning a fortune on a deluge of fraudulent ads, documents show

But the company only bans advertisers if its automated systems predict the marketers are at least 95% certain to be committing fraud, the documents show. If the company is less certain – but still believes the advertiser is a likely scammer – Meta charges higher ad rates as a penalty, according to the documents. The idea is to dissuade suspect advertisers from placing ads.

🔖 Servo: A new web engine written in Rust

Detailed description slide by slide of my Servo talk at GOSIM Hangzhou 2025.

🔖 DuckDB Internals (CMU Advanced Databases / Spring 2023)

15-721 Advanced Database Systems (Spring 2023) Carnegie Mellon University.

🔖 The Raft Consensus Algorithm

Raft is a consensus algorithm that is designed to be easy to understand. It’s equivalent to Paxos in fault-tolerance and performance. The difference is that it’s decomposed into relatively independent subproblems, and it cleanly addresses all major pieces needed for practical systems. We hope Raft will make consensus available to a wider audience, and that this wider audience will be able to develop a variety of higher quality consensus-based systems than are available today.

🔖 anyproto/any-sync

any-sync is an open-source protocol designed for the post-cloud era, enabling high-speed, peer-to-peer synchronization of encrypted communication channels (spaces). It provides a communication layer for building private, decentralized applications offering unparalleled control, privacy, and performance.

🔖 Indexing coffee with Notion

This article is a reinterpretation of an article I wrote in 2021 on my former site. I tried to focus on the essentials, and it outlines the set of Notion pages I created to index my coffee consumption and attempt to build an intuition about understanding my tastes. In general, I think that building a knowledge base is a good practice when trying to explore a discipline. Being far from an expert in both coffee and Notion, it’s likely that much of what I describe in this article may seem naïve! To sum up, this article will present how I set up an infrastructure to index the coffees I taste, with the goal of providing precise metrics to help characterize my preferences, using the Notion tool, while also sharing some techniques and tricks I learned during the creation of this system.

🔖 Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models

We present evidence that adversarial poetry functions as a universal single-turn jailbreak technique for Large Language Models (LLMs). Across 25 frontier proprietary and open-weight models, curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90%. Mapping prompts to MLCommons and EU CoP risk taxonomies shows that poetic attacks transfer across CBRN, manipulation, cyber-offence, and loss-of-control domains. Converting 1,200 MLCommons harmful prompts into verse via a standardized meta-prompt produced ASRs up to 18 times higher than their prose baselines. Outputs are evaluated using an ensemble of 3 open-weight LLM judges, whose binary safety assessments were validated on a stratified human-labeled subset. Poetic framing achieved an average jailbreak success rate of 62% for hand-crafted poems and approximately 43% for meta-prompt conversions (compared to non-poetic baselines), substantially outperforming non-poetic baselines and revealing a systematic vulnerability across model families and safety training approaches. These findings demonstrate that stylistic variation alone can circumvent contemporary safety mechanisms, suggesting fundamental limitations in current alignment methods and evaluation protocols.

🔖 FediGroups.social

The concept is simple: If you mention a FediGroup in one of your posts, it will automatically be shared with everyone who follows the group.

🔖 From LOD to LOUD: making data usable

Slides from SWIB 2018 about the importance of JSON to Linked Data and the idea of Linked Open Usable Data (LOUD).

🔖 Monitoring machine learning models for bot detection

To train a model for the Internet is to train a model against a moving target. Anyone can train a model on static data and achieve great results — so long as the input does not change. Building a model that generalizes into the future, with new threats, browsers, and bots is a more difficult task. Machine learning monitoring is an important part of the story because it provides confidence that our models continue to generalize, using a rigorous and repeatable process.

🔖 Cloudflare outage on November 18, 2025

Cloudflare’s Bot Management includes, among other systems, a machine learning model that we use to generate bot scores for every request traversing our network. Our customers use bot scores to control which bots are allowed to access their sites — or not.

The model takes as input a “feature” configuration file. A feature, in this context, is an individual trait used by the machine learning model to make a prediction about whether the request was automated or not. The feature configuration file is a collection of individual features.

This feature file is refreshed every few minutes and published to our entire network and allows us to react to variations in traffic flows across the Internet. It allows us to react to new types of bots and new bot attacks. So it’s critical that it is rolled out frequently and rapidly as bad actors change their tactics quickly.

A change in our underlying ClickHouse query behaviour (explained below) that generates this file caused it to have a large number of duplicate “feature” rows. This changed the size of the previously fixed-size feature configuration file, causing the bots module to trigger an error.

🔖 Cost Per Article: A transparent view of what it takes to support rigorous, accessible, and sustainable science.

AIP Publishing’s mission is to advance, promote, and serve the physical sciences for the benefit of humanity. We believe that openness builds trust: in research, in publishing, and in the scientific enterprise itself.

Our 2024 cost per article represents the real investment it takes to publish a single, peer-reviewed article, based on 2024 operations. This figure reflects the full scope of services that enable trustworthy science, from editorial oversight to digital preservation.

The cost per article in 2024 was $2,700

🔖 The Laugh of the Medusa

In the essay, Cixous issues an ultimatum: that women can either read and choose to stay trapped in their own bodies by a language that does not allow them to express themselves, or they can use the body as a way to communicate. She describes a writing style, écriture féminine, that she says attempts to move outside of the conventional rules found in patriarchal systems. She argues that écriture feminine allows women to address their needs by building strong self-narratives and identity. This text is situated in a history of feminist conversations that separated women because of their gender especially in terms of authorship.[1] The “Laugh of the Medusa” addresses this rhetoric, writing on individuality and commanding women to use writing and the body as sources of power and inspiration.

🔖 It’s your fault my laptop knows where I am

So, that’s what Apple, Google, and Microsoft devices began doing. The location services of their products, by default, started aggregating the SSIDs and BSSIDs of Wi-Fi hotspots they could see (and their locations) and logging them for others’ devices to use for more accurate location services. And… that’s more or less the same thing that modern devices use today. When Chrome tells me that a website would like to use my location, and I allow it, the list of the surrounding hotspots will be sent to Google — which, because tens of thousands of people with GPS-enabled devices have also pinged the networks, allows my computer to obtain an extremely accurate estimation on where I am. So, thank you, everybody…?

🔖 Web Archive Shapes and Schema

Web archiving is the endeavor of preserving the web. The web is born-digital, multimedial, consist of hyper text, and distributed resources. As such it is fundamentally different to other media types, that are traditionally collected in archives and libraries. To govern the archival process, a data model is required that provides the flexibility to express and describe the web archival materials properties. The Resource Description Framework (RDF) provides such a data model build on top and into the web technology stack.

🔖 Linked Open Usable Data for Cultural Heritage: Community Building and Semantic Interoperability in Practice

This paper presents an extended transcript of a talk given online on 18 November 2025 for the 17th Semantic Web in Libraries Conference (SWIB25). It shares key findings from my PhD thesis on Linked Open Usable Data (LOUD) for cultural heritage. My research examined how LOUD specifications like IIIF APIs and Linked Art fostered collaborative knowledge creation, focusing on implementations in both the Participatory Knowledge Practices in Analogue and Digital Image Archives (PIA) project and Yale’s LUX platform. Using a framework based on Actor-Network Theory (ANT), the analysis revealed three critical dimensions. First, sustainable development required continuous engagement beyond implementation, with community-led practices providing the socio-technical foundation for specification maintenance. Second, demographic homogeneity perpetuated biases that marginalised diverse perspectives, requiring the transformation of inclusion frameworks. Third, LOUD improved the discoverability of heritage data while requiring investment in accessibility paradigms that acknowledged technological differences. The research demonstrates that LOUD methodologies foster collaborative knowledge production through community engagement, confront power dynamics in inclusion frameworks, and provide mechanisms for democratising heritage access while accounting for technological disparities.

🔖 The Data Center Resistance Has Arrived

Georgia has become a hot spot for data center development over the past few years: Some research indicates it’s one of the fastest-growing markets for data center development in the country (thanks, in part, to some generous tax breaks). It’s also now a nexus for organizing against those same data centers. Community opposition to data centers, a new report finds, is on the rise across the country. And red states, including Georgia and Indiana, are leading this wave of bipartisan opposition

🔖 Bubble or Nothing

Should economic conditions in the tech sector sour, the burgeoning artificial intelligence (AI) boom may evaporate—and, with it, the economic activity associated with the boom in data center development.

Policymakers concerned about the deployment of clean energy and compute-focused infrastructure over the long term need a framework for managing the uncertainty in this sector’s investment landscape—and for understanding the local and regional impacts of a market correction that strands data centers and their energy projects. This framework requires understanding how a potential downward market correction in the tech sector might occur and, if so, how to sustain investment in critical energy infrastructure assets during potentially recessionary conditions.

🔖 Haecceity

Haecceity (/hɛkˈsiːɪti, hiːk-/; from the Latin haecceitas, ‘thisness’) is a term from medieval scholastic philosophy, first coined by followers of Duns Scotus to denote a concept that he seems to have originated: the irreducible determination of a thing that makes it this particular thing. Haecceity is a person’s or object’s thisness, the individualising difference between the concept “a person” and the concept “Socrates” (i.e., a specific person). In modern philosophy of physics, it is sometimes referred to as primitive thisness.

🔖 Ears To The Ground

For the biggest artists to the most underground, field recordings have become the vital spark of electronic music. Whether documenting nature, sampling the city or capturing the atmosphere of archaeological sites, musicians are using found sounds to make sense of our world. Ears To The Ground explores the relationship between electronics, landscape and field recordings in the UK, Ireland and around the globe, discovering how producers and artists evoke the natural world, history and folklore through sampled sounds.

🔖 Bumping Into a Chair While Humming: Sounds of the Everyday, Listening, and the Potential of the Personal

Bumping Into a Chair While Humming explores the sonic potential in everyday objects, spaces, and interactions - the importance of recognizing happy accidents and using the tools at your disposal toward creative ends. It concentrates on how to create a personal soundscape by searching for the moments in one’s immediate environment that resonate for the individual, while editing, arranging, and completing work. The author, Ezekiel Honig, imparts clues into his favored music production processes, but the book is more focused on the practice of listening itself, and how that benefits one’s art, and life in general. It plays with ideas in the creative process and how to use them, through anecdotal qualities and illustrations of hypothetical moments ranging from the associations we have with inanimate objects, utilizing different types of spaces, and experimenting with rhythm.

The book is punctuated by, and highlighted with, illustrations by Asli Senel Smith – complex line drawings that abstractly define the subjects of each chapter, concretizing the concepts on the page and capturing the introspective, yet expansive tone.

🔖 Who needs Graphviz when you can build it yourself?

We recently overhauled our internal tools for visualizing the compilation of JavaScript and WebAssembly. When SpiderMonkey’s optimizing compiler, Ion, is active, we can now produce interactive graphs showing exactly how functions are processed and optimized.

We are not the first to visualize our compiler’s internal graphs, of course, nor the first to make them interactive. But I was not satisfied with the output of common tools like Graphviz or Mermaid, so I decided to create a layout algorithm specifically tailored to our needs. The resulting algorithm is simple, fast, produces surprisingly high-quality output, and can be implemented in less than a thousand lines of code. The purpose of this article is to walk you through this algorithm and the design concepts behind it.

🔖 Plunderphonics by Matthew Blackwell

In Plunderphonics, Matthew Blackwell tells the story of a group of musicians who advocated for changes to the copyright system by deploying unlicensed samples in their recordings. The composer John Oswald, who coined the genre term “plunderphonics,” was threatened with legal action by the Canadian Recording Industry Association on behalf of Michael Jackson. The Bay Area group Negativland was sued by Island Records on behalf of U2 for their parody of the band. These artists attracted media attention to their cause in a bid to expand fair use protections. Later, the Australian band the Avalanches encountered the limitations of the music licensing system during the release of their debut album, having to drop several samples that could not be successfully cleared. Finally, American DJ and producer Girl Talk released a series of albums featuring hundreds of uncleared samples and successfully avoided lawsuits by publicly arguing a fair use defense.

🔖 OpenAlex API Responses Notebook

TL;DR: there are quite a few undocumented fields returned by the API, and fields that have a different structure compared to the docs. I made a quick notebook with python dataclasses to test these issues which you can run yourself from your browser here as an app (as shown in the screenshot), or here as a notebook w/ editable source code.