|
📖 Read In Depth
|
Andrew Curran: Anthropic May Have Had An Architectural Breakthrough!
Rumors circulating that Anthropic completed its largest-ever training run with results significantly exceeding both internal expectations and scaling law predictions. This is unverified but the discussion includes credible-seeming details about architectural changes. Worth reading the thread critically — if true, it signals a potential step change in capability from Anthropic.
reddit/r/singularity
|
The first 40 months of the AI era
A thoughtful retrospective on the first 40 months of the AI era, grappling honestly with the question of whether productivity improvements are real and how large they actually are. The HN comment from 'theteapot' captures the core tension well: genuine uncertainty about impact despite obvious capability gains. Good for calibrating your own priors.
hn/Best Stories
|
AMD's Ryzen 9 9950X3D2 Dual Edition crams 208MB of cache into a single chip
AMD's new 9950X3D2 stacks dual 3D V-Cache dies for 208MB total L3 on a single package — nearly double the previous generation. For ML inference and simulation workloads where memory bandwidth and latency dominate, this cache architecture matters more than raw clock speed. Relevant to the build-vs-buy calculus for local inference.
hn/Best Stories
|
Local Opposition Is Slowing A.I. Data Centers. Wall Street Has Noticed.
Local opposition to AI data centers — noise, water use, grid load, property concerns — is materially slowing construction timelines, and Wall Street is pricing in the risk. The piece examines the structural constraint: hyperscalers made capacity promises that depend on build rates now being blocked by community resistance. Relevant to understanding the real bottleneck in AI infrastructure.
nyt/Business
|
Linux is an interpreter
An argument that Linux (and Unix-like kernels generally) are better understood as interpreters of a bytecode-like ABI than as monolithic programs — reframing how system calls, ELF loading, and process isolation work conceptually. The kind of bottom-up reframing that's useful if you like understanding systems from first principles.
hn/Best Stories
|
|
🎬 Check It Out
|
Spanish legislation as a Git repo
A developer converted all 8,642 Spanish state laws into a Git repo where each reform is a real commit with the historical date — 27,866 commits total. Available now on GitHub. The builder's HN comment explains the pipeline; the French commenter notes France goes further with formally-proved law via Catala. A genuinely clever application of version control to a non-software domain.
hn/Best Stories
|
|
âš¡ FYI
|
Further human + AI + proof assistant work on Knuth's "Claude Cycles" problem
Knuth's 'Claude Cycles' combinatorics problem — which Knuth himself posed and couldn't solve — has now been fully solved using a human + LLM + formal proof assistant collaboration. The HN thread has the full context including links to the original problem and the solution. A concrete data point on what human-AI mathematical collaboration can now achieve.
hn/Best Stories
|
Global Food Supply Faces a Dangerous Bottleneck as Iran War Persists
The Iran war is disrupting fertilizer supply chains — much of global potash and ammonia transit moves through or near the Persian Gulf — pushing food prices up worldwide. A second-order effect of Hormuz closure that isn't getting as much attention as oil prices but could be more persistent.
nyt/Business
|
How the Iran War Is Costing the Economy Its Buffers
Comprehensive DealBook breakdown of the Iran war's economic toll: S&P down ~9% from January highs, one forecast puts oil at $200/barrel, inflation buffers eroding. The clearest single-article summary of the macro situation if you want the full picture in one read.
nyt/Business
|
I decompiled the White House's new app
Someone decompiled the new White House app and found the code largely unremarkable — but the HN thread is worth skimming because commenters, including one who independently downloaded and verified, pushed back on AI-assisted analysis errors in the original post. A useful case study in the limits of LLM-assisted reverse engineering.
hn/Best Stories
|
Overestimation of microplastics potentially caused by scientists' gloves
University of Michigan study finds that nitrile and latex lab gloves shed microplastic particles that contaminate samples, potentially causing significant overestimation of environmental microplastics concentrations. The HN top comment draws the apt parallel to the European forensic DNA contamination scandal. A methodological artifact that could require re-evaluation of a large body of research.
hn/Best Stories
|