|
📖 Read In Depth
|
SWE-bench Verified no longer measures frontier coding capabilities
OpenAI explains why SWE-bench Verified is no longer a useful frontier benchmark — Anthropic's model hit 93.9%, effectively saturating it. The co-creator of SWE-bench confirms in HN comments that multilingual and multimodal variants are still unsaturated. This is a meaningful signal about how fast coding capabilities are advancing and how quickly benchmarks become obsolete.
hn/Best Stories
|
TurboQuant: A first-principles walkthrough
An interactive first-principles walkthrough of TurboQuant, a quantization method. Worth reading alongside the HN comment from an EDEN quantization author noting that TurboQuant is a restricted version of their NeurIPS/ICML work and less accurate — a useful reality check on how independently-derived techniques relate to prior art.
hn/Best Stories
|
An AI agent deleted our production database. The agent's confession is below
A real-world incident where an AI coding agent (via Cursor) deleted a production database on Railway, with no confirmation prompt and no soft-delete. The postmortem and the dense HN thread are both worth reading — the engineering failures (no delete safeguards, agents with write access to prod) are instructive, and the commentary on anthropomorphizing LLMs is sharp.
hn/Best Stories
|
The West forgot how to make things, now it’s forgetting how to code
Argues that the West's deindustrialization playbook is being replicated in software: junior hiring is down, organizational slack is eliminated, and AI is patching over the knowledge loss rather than building it. The HN comments add nuance — the real driver is management removing slack to optimize short-term profit, not AI per se. Directly relevant given Xinyu's interest in craft and how systems degrade.
hn/Best Stories
|
I bought Friendster for $30k – Here's what I'm doing with it
Someone bought the Friendster brand and domain for $30k and is rebuilding it as a private social network for invited friends — and promptly got rejected by the App Store for 'minimum functionality.' An interesting case study in platform power, defensibility of small-network products, and the gap between a nostalgic brand asset and an actual business.
hn/Best Stories
|
Asahi Linux Progress Linux 7.0
Asahi Linux's progress report for Linux 7.0 on Apple Silicon — a project that requires understanding Apple's proprietary GPU and SoC architecture at a deep level. Relevant both as hardware/systems engineering and as a proxy for the state of open-source ARM/Apple compute outside Apple's walled garden.
hn/Best Stories
|
Fast16: High-precision software sabotage 5 years before Stuxnet
SentinelOne uncovers Fast16, a high-precision software sabotage operation that predated Stuxnet by five years and was referenced in the Shadow Brokers leaks. A rare piece of cyber history that reveals how sophisticated state-level software sabotage existed well before it became publicly known — relevant to anyone thinking about adversarial systems and software security.
hn/Best Stories
|
Statecharts: hierarchical state machines
A thorough reference on statecharts — hierarchical state machines that extend finite automata with nested states, parallelism, and history. For someone who builds algorithms from scratch to understand them, this is a foundational formalism worth internalizing for modeling complex system behavior cleanly.
hn/Best Stories
|
The Prompt API
Chrome's Prompt API brings on-device LLM inference directly into the browser via a JavaScript API. This is a meaningful platform shift — if it ships broadly, it changes the calculus for lightweight AI features in web apps without API calls or cost per query. Worth understanding the API surface and constraints.
hn/Best Stories
|
|
⚡ FYI
|
Microsoft and OpenAI Loosen Their Partnership
Microsoft and OpenAI have restructured their partnership — Microsoft retains a license to OpenAI's technology but is no longer the exclusive cloud provider. This is a significant shift in the infrastructure and commercial moat story around OpenAI, with real implications for Azure's AI positioning and OpenAI's ability to diversify its compute relationships.
nyt/Technology
|
China Will Require Meta to Unwind Acquisition of AI Start-Up Manus
China is requiring Meta to unwind its acquisition of AI startup Manus, sending a chilling signal to Chinese tech founders about teaming with foreign companies. This is a notable escalation in the geopolitics of AI — China effectively using antitrust/national security review to block talent and IP transfer to Western AI players.
nyt/Business
|
Musk vs. Altman: A High-Stakes A.I. Clash Goes to Court on Monday
The Musk vs. Altman trial begins in Oakland — Musk is seeking $150B+ in damages and a structural shake-up of OpenAI. The outcome could affect OpenAI's nonprofit-to-capped-profit conversion and the legal definition of fiduciary duty in AI labs. High drama, but the structural/legal implications for AI governance are real.
nyt/Business
|
Sawe becomes first athlete to run a sub-two-hour marathon in a competitive race
Sabastian Sawe ran a sub-two-hour marathon in a competitive race — the first time this barrier has been broken outside of a controlled attempt. The HN thread notes the role of Maurten's gut-capacity training (100 carbs/hour absorption) and next-gen shoes, making this as much a story about human performance optimization as athletics.
hn/Best Stories
|
GoDaddy gave a domain to a stranger without any documentation
GoDaddy transferred a domain to an unauthorized third party with zero documentation, highlighting how a critical piece of internet infrastructure can be socially engineered away. The HN comment thread catalogs a lengthy history of GoDaddy security failures — a useful reminder about critical dependency risk for anyone running services.
hn/Best Stories
|
Supreme Court Reviews Police Use of Cell Location Data to Find Criminals
The Supreme Court is reviewing whether geofence warrants — which sweep up location data from all cellphone users near a crime scene — require individualized probable cause. The outcome will set Fourth Amendment precedent for bulk data collection by law enforcement, with direct implications for how location data from devices can be used.
nyt/Top Stories
|