Weekly Bookmarks

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

🔖

Talk Talk - Live at Montreux 1986

From Aquarium Drunkard:

By July 1986, Talk Talk were still a functioning live unit touring behind The Colour of Spring. But something had already shifted as evidenced by this set from that summer’s Montreux Jazz Festival. Listen closely and you can hear the architecture beginning to loosen: tempos breathe, arrangements open, and familiar material begins to drift toward something less fixed, less performative.

This would be their final tour. Within a year, Mark Hollis and company would retreat into the studio to begin work on Spirit of Eden, a record that all but rejects the idea of live translation. As such, this Montreux performance exists as a kind of threshold document, one that captures the band onstage one last time before the music folds inward on itself.

🔖

The West Forgot How to Make Things. Now It’s Forgetting How to Code

The skills you need to be effective now are different. Technical expertise alone isn’t enough anymore. You need people who can take ownership, communicate tradeoffs, push back on bad suggestions from a machine that sounds very confident. Leadership qualities. Our last hiring round tells you how rare that is: 2,253 candidates, 2,069 disqualified, 4 hired. A 0.18% conversion rate. The combination of technical skill and the judgment to know when the AI is wrong barely exists in the market anymore.

🔖 There Will

Be a Scientific Theory of Deep Learning

In this paper, we make the case that a scientific theory of deep learning is emerging. By this we mean a theory which characterizes important properties and statistics of the training process, hidden representations, final weights, and performance of neural networks. We pull together major strands of ongoing research in deep learning theory and identify five growing bodies of work that point toward such a theory: (a) solvable idealized settings that provide intuition for learning dynamics in realistic systems; (b) tractable limits that reveal insights into fundamental learning phenomena; (c) simple mathematical laws that capture important macroscopic observables; (d) theories of hyperparameters that disentangle them from the rest of the training process, leaving simpler systems behind; and (e) universal behaviors shared across systems and settings which clarify which phenomena call for explanation.

🔖 EHRAG:

Bridging Semantic Gaps in Lightweight GraphRAG via Hybrid Hypergraph Construction and Retrieval

Graph-based Retrieval-Augmented Generation (GraphRAG) enhances LLMs by structuring corpus into graphs to facilitate multi-hop reasoning. While recent lightweight approaches reduce indexing costs by leveraging Named Entity Recognition (NER), they rely strictly on structural co-occurrence, failing to capture latent semantic connections between disjoint entities. To address this, we propose EHRAG, a lightweight RAG framework that constructs a hypergraph capturing both structure and semantic level relationships, employing a hybrid structural-semantic retrieval mechanism. Specifically, EHRAG constructs structural hyperedges based on sentence-level co-occurrence with lightweight entity extraction and semantic hyperedges by clustering entity text embeddings, ensuring the hypergraph encompasses both structural and semantic information. For retrieval, EHRAG performs a structure-semantic hybrid diffusion with topic-aware scoring and personalized pagerank (PPR) refinement to identify the top-k relevant documents. Experiments on four datasets show that EHRAG outperforms state-of-the-art baselines while maintaining linear indexing complexity and zero token consumption for construction.

🔖

The Night Manager (British TV series)

The Night Manager is a British spy thriller television serial based on the 1993 novel by John le Carré and adapted by David Farr. The six-part first series, directed by Susanne Bier and starring Tom Hiddleston, Hugh Laurie, Olivia Colman, Tom Hollander, David Harewood and Elizabeth Debicki, began broadcasting on BBC One on 21 February 2016.

🔖

What Will It Take to Get A.I. Out of Schools?

Immordino-Yang told me that the ultimate goal of any school assignment is not the finished project itself but the experience of having done it—an experience that A.I. tools are intended to abbreviate or obviate. With their prettifying intrusions and impatient, lurking presence, they block and reroute a young person’s natural, gradual progression toward cognitive maturity, “especially one who is still developing the neuropsychological substrate for creating narratives and thinking through arguments over time,” Immordino-Yang said. “It’s a fragile process, and it’s being interrupted.”

🔖

iocaine

This software is not made for making the Crawlers go away. It is an aggressive defense mechanism that tries its best to take the blunt of the assault, serve them garbage, and keep them off of upstream resources. Even though a lot of work went into making iocaine efficient, and nigh invisible for the legit visitor, it is an aggressive defender nevertheless, and will require a few resources - a whole lot less than if you’d let the Crawlers run rampant, though.

Before you deploy it, be sure you understand that iocaine does not make the bots go away. It tries to poison them, so they’d go away forever in the long run. If you’re looking for a way to return the favour, to “reward” these crawlers for their relentless assault, this is the tool you’re looking for.

🔖

AI as a Fascist Artifact

“AI” is being introduced increasingly into government processes: “AI” is promised to bring more efficiency into the administration, is supposed to “reduce bureaucracy”. But bureaucracy is not just an annoyance but one of the central tools that democratic societies have established to realize the core idea of democracy: Transparency in the application of power in order to be able to control said power. Democracy is not just about voting but about ensuring that all power – especially by the state – is used in accordance with the law and in a fair way. Stochastic “AI” systems break that promise. The “AI” just says that you do not get the support you need. No idea why, might be a bug or a deeply racist training data set or something else. Nobody knows. Now it is on you to prove that you are in the right, it is on you to fight for your right because the processes that were supposed to protect your rights are hollowed out in order to make them faster: We are forcing marginalized, disenfranchised people to fight against a black box trained on the data that already contains their disenfranchisement

🔖 You’re about to

feel the AI money squeeze

Investors have poured hundreds of billions of dollars into companies like OpenAI and Anthropic to help them scale and build out their compute. Now, they’re expecting returns. After years of offering cheap or totally free access to advanced AI systems, the bill is starting to come due — and downstream, users are beginning to feel the pinch.

🔖 RustFS

Instantly replace MinIO & S3. Zero GC, maximum throughput.

🔖

Finishing Things

Exactly! Left behind? You can’t leave me behind fast enough. I’ve never wanted to be left behind so bad in my life. I’m utterly incapable of FOMO about this stuff. Do I have vague future concerns about my career and what must (surely) be a coming economic crash? Sure, but there is absolutely nothing that has convinced me that I’m “missing out” on anything.

At this point, all efforts by boosters and sloppologists just make me feel more defiant.

🔖

Go Ahead And Use AI. It Will Only Help Me Dominate You.

The tepid, conformist nature of your AI-assisted prose will only make my unexpected bons mots stand out more sharply. While you lean on a technological crutch of grammatical mediocrity to drag your essays over the finish line, I’ll be metaphorically zipping past you on my “magic carpet” of words emerging directly from my own declining and unpredictable brain. Over time, the intellectual box into which AI has seduced your creative process will suffocate you, leaving your bereft readers little choice but to drift into my subscription base.

🔖

Friends Don’t Let Friends Use Ollama

Ollama gained traction by being the first easy llama.cpp wrapper, then spent years dodging attribution, misleading users, and pivoting to cloud, all while riding VC money earned on someone else’s engine. Here’s the full history, and why the alternatives are better