|
📖 Read In Depth
|
Claude Code is steganographically marking requests
Researchers discovered that Claude Code is embedding steganographic markers in its requests — hidden signals invisible to users but detectable by Anthropic's infrastructure. This is a significant finding about how AI coding tools may be covertly tracking or watermarking usage, with real implications for trust, transparency, and the mechanics of how these systems operate under the hood.
hn/Best Stories
|
Claude Sonnet 5
Anthropic released Claude Sonnet 5 with high-res vision, a new tokenizer, and high-effort mode — positioned as near-Opus performance at Sonnet pricing. Community benchmarks suggest it may not actually exceed Opus 4.8 on coding tasks, and Anthropic confirmed some coding/debugging will route back to Opus 4.8, raising questions about whether this is a genuine capability jump or a cost-optimization play.
hn/Best Stories
|
Claude Science
Anthropic launched Claude Science, a research-focused workflow built on Opus 4.8 that integrates PubMed, Jupyter, R, cluster terminals, and scientific databases into a single artifact-first, auditable environment. This is essentially Claude Code for wet-lab and computational researchers — worth understanding if you think about where AI-assisted research pipelines are heading.
hn/Best Stories
|
The labor share of income in the US is at its lowest post-war level
The NY Fed documents that the labor share of US income is at its lowest post-war level. This is the structural backdrop behind AI-driven productivity debates — not just a statistic, but a signal about who's capturing the gains from automation and what the political economy of AI deployment will look like.
hn/Best Stories
|
Godot will no longer accept AI-authored code contributions
Godot's maintainers are banning AI-authored code contributions on the grounds that heavy AI users don't understand their own code well enough to debug or maintain it. This is a direct challenge to the 'AI accelerates engineering' narrative from people who actually review code quality at scale — substantive for anyone thinking about what software craftsmanship means in the AI era.
hn/Best Stories
|
We Are the Last People Who Know How It Works
An essay arguing that the current generation of engineers may be the last who understand systems at a deep level, as AI-assisted development abstracts away foundational knowledge. Directly relevant to Xinyu's philosophy of building from scratch to understand deeply — likely to either sharpen or challenge that conviction.
hn/Best Stories
|
The South Korean Mine at the Center of America’s Tungsten Push
South Korea is reviving a tungsten mine as part of a broader push to break China's grip on a metal critical to weapons manufacturing, semiconductors, and tooling. This is the supply-chain geopolitics story that rarely gets told — tungsten sits upstream of chip fabrication equipment and defense systems alike.
nyt/Business
|
Anthropic Won a Reprieve From the Trump Administration. Is It Enough?
DealBook digs into the broader implications of Anthropic's export control reprieve: Silicon Valley is now operating under a regime where the White House can effectively blacklist AI models, and the 'alignment with US government' framing in Lutnick's letter signals a new kind of political capture risk for AI labs. More nuanced than the straight news piece.
nyt/Business
|
County with 37 Data Centers Asks Schools to 'Conserve Electricity'
A Virginia county with 37 data centers is asking schools to conserve electricity — a vivid illustration of how AI infrastructure is creating real energy externalities that fall on public institutions. The infrastructure cost story is usually told in watts and dollars; this tells it in school emails.
hn/Best Stories
|
Asahi Linux 7.1 Progress Report
Asahi Linux's 7.1 progress report covers low-level work on Apple Silicon Linux support — GPU drivers, kernel patches, and hardware bring-up. Relevant for anyone interested in the craft of OS/hardware development and the ongoing project of understanding Apple's tightly controlled silicon from the outside.
hn/Best Stories
|
|
⚡ FYI
|
Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5
The Trump administration lifted export controls on Anthropic's Fable 5 and Mythos 5 models, restoring global access after a weeks-long standoff. Commerce Secretary Lutnick framed it as a national security review; the deeper story is that Washington is now explicitly vetting which AI models can be deployed internationally — a new regulatory reality for frontier labs.
hn/Best Stories
|
CERN bids farewell to the LHC and enters Long Shutdown 3
CERN has shut down the LHC for Long Shutdown 3, a multi-year upgrade period before the High-Luminosity LHC era begins. A genuine milestone in physics infrastructure — the machine that found the Higgs boson is going offline for a significant overhaul.
hn/Best Stories
|
The first early human eggs from stem cells - Conception | Advancing the Future of Fertility
A biotech company reports producing early human eggs from stem cells in what could be a foundational step toward in vitro gametogenesis — lab-grown human reproductive cells. If it holds up, this is a major biology story with profound implications for fertility, genetics, and reproductive medicine.
reddit/r/singularity
|
As Companies Race for Cheaper A.I. Options, This Start-Up Pitches a Solution
Together AI, which runs inference on open-source models, crossed $8B valuation as enterprises seek cheaper alternatives to proprietary frontier models. The 'build vs. buy' calculus is shifting — and companies that abstract inference costs over open weights are carving out real defensible positions.
nyt/Business
|