|
📖 Read In Depth
|
Hardware Attestation as Monopoly Enabler
Hardware attestation — where devices cryptographically prove their software stack to remote services — is increasingly being used as a monopoly mechanism. Google and Apple can gatekeep which devices can access services, making open/alternative platforms effectively second-class. The HN thread is rich with implications for EU digital identity, open-source software, and platform lock-in.
hn/Best Stories
|
Mythos Finds a Curl Vulnerability
Daniel Stenberg (curl's author) breaks down how the Mythos AI system found a real vulnerability in curl — and critically assesses whether the hype is warranted. His conclusion: the vulnerability is real but Mythos doesn't obviously outperform existing tools; the marketing exceeds the results. Essential reading for anyone evaluating AI-assisted security research claims.
hn/Best Stories
|
I'm going back to writing code by hand
A developer argues for returning to hand-written code after extended AI-assisted development — citing degraded understanding, harder debugging, and skill atrophy. Directly relevant to Xinyu's philosophy of building from scratch to understand deeply; this is the practitioner's counterargument to the AI-coding wave.
hn/Best Stories
|
Local AI needs to be the norm
Argues that cloud-dependent AI creates unacceptable privacy and sovereignty risks, and that local inference should become the default. The HN thread has substantive discussion on the hardware trajectory — M4 MacBooks and Strix Halo as credible local inference platforms — making this a good read alongside the M4 local model benchmarks article.
hn/Best Stories
|
Running local models on an M4 with 24GB memory
Practical benchmarks running local LLMs on Apple M4 with 24GB unified memory — covering throughput, model sizes that fit comfortably, and real-world usability. Directly useful for an ML engineer evaluating local inference as a serious option versus API calls.
hn/Best Stories
|
An AI coding agent, used to write code, needs to reduce your maintenance costs
Makes the underappreciated argument that AI coding agents should be evaluated on long-term maintenance cost reduction, not just feature velocity. Code that ships fast but is hard to test, understand, or modify is a liability — the piece offers a framework for thinking about AI coding ROI that goes beyond hype metrics.
hn/Best Stories
|
Maryland citizens hit with $2B power grid upgrade for out-of-state AI
Maryland ratepayers are being billed $2B for grid upgrades primarily benefiting out-of-state AI data centers — a concrete case study in how AI infrastructure costs are being externalized onto local populations. Connects AI compute buildout to energy policy, incentive structures, and regulatory capture.
hn/Best Stories
|
A Single Infusion Could Suppress H.I.V. for Years, Study Suggests
A small but striking study suggests a single infusion using broadly neutralizing antibodies (bnAbs) can suppress HIV for years — building on immunotherapy approaches that have already achieved functional cures in some blood cancer patients. Early-stage but conceptually significant if it scales: shifting HIV from daily pills to rare interventions.
nyt/Top Stories
|
|
🎬 Check It Out
|
Spectacle vs subjectivity: Poor Things and The Bride! on womanhood
A substantive critical essay comparing Poor Things and The Bride! as contrasting approaches to feminist filmmaking — argues Poor Things routes female 'freedom' through a male gaze while The Bride! gives its protagonist genuine interiority. The Bride! is a 2025 film currently in circulation; worth watching before reading this discussion if you haven't seen it.
reddit/r/TrueFilm
|
|
⚡ FYI
|
After Shopify and Google said that 50% and 75% of their code is AI-generated, it’s now Airbnb’s turn to say that 60% of its codebase is also AI-generated. Moreover, Airbnb's CEO says that even managers are programming with Claude Code.
Airbnb's CEO says 60% of new code is AI-generated and even managers are using Claude Code — joining Shopify (50%) and Google (75%) in this wave of announcements. The trend is real and accelerating; the question of what this means for engineering headcount and skill distribution is increasingly urgent.
reddit/r/singularity
|
Google Says Criminal Hackers Used A.I. to Find a Major Software Flaw
Google reports the first confirmed case of criminal hackers using AI to autonomously discover an unknown (zero-day) vulnerability in software. This is a qualitative shift — AI as an offensive security tool in the wild, not just in research settings. The 'taste of what's to come' framing from security researchers is worth taking seriously.
nyt/Top Stories
|
MIT FINGERS-7B: First Multi-Omics AI Model for Alzheimer’s Prevention
MIT released FINGERS-7B, a multi-omics foundation model trained on 8 trillion tokens from 30k patients across genetics, biomarkers, and lifestyle data for Alzheimer's risk prediction. Claims early detection years before symptoms — model is public but gated. A notable example of large-scale biological foundation models moving into clinical-adjacent territory.
reddit/r/singularity
|
Scientists Press Congress on Dismissal of National Science Foundation Board and Research Funding
Scientists are pressing Congress on the White House's gutting of NSF — slowed grants, dismissed board members, and reduced research funding. The explicit concern raised: ceding scientific leadership to China. Directly relevant to anyone thinking about US long-term competitiveness in AI and foundational research.
nyt/Top Stories
|
ICLR 2026 Contributions by Country/Institution
Breakdown of ICLR 2026 paper contributions by country and institution — useful as a proxy for where ML research talent and output is concentrated. The US-China comparison is particularly relevant given NSF funding debates and compute geopolitics.
reddit/r/singularity
|
As Coal Rebounds, More Toxic Mercury Is in the Air
Coal-burning power plants released more mercury in 2025, reversing a decade-long downward trend — a direct consequence of rolling back emissions regulations. Mercury is a potent neurotoxin that bioaccumulates. The AI data center energy demand story and coal rebound are increasingly connected threads.
nyt/Top Stories
|