macOS Container Machines

System Card [pdf]: <a href="https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf" rel="nofollow">https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c3...</a>

Claude Fable 5

<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=K5LAFEjTlBA" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=K5LAFEjTlBA</a>

Rich Sutton on AI creativity and discovery

Upcoming breaking changes for npm v12

German ruling declares Google liable for false answers in AI Overviews

RIP software hackathons. Long live the hardware hackathon

The oldest surviving animated feature film at 100

<a href="https:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20260609200156&#x2F;https:&#x2F;&#x2F;aarushgupta.io&#x2F;posts&#x2F;kan-fpga&#x2F;" rel="nofollow">https:&#x2F;&#x2F;web.archive.org&#x2F;web&#x2F;20260609200156&#x2F;https:&#x2F;&#x2F;aarushgup...</a>

Ultrafast machine learning on FPGAs via Kolmogorov-Arnold Networks

More Molly Guards

Related: https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;Jun&#x2F;10&#x2F;if-claude-fable-stops-helping-you&#x2F;

If Claude Fable stops helping you, you'll never know

What it feels like to work with Mythos

CEOs who think AI replaces their employees are just bad CEOs

Value Numbering

OpenCV 5 Is Here: The Biggest Leap in Years for Computer Vision

Test-case reducers are underappreciated debugging tools

Exif Smuggling (2025)

Grit: Rewriting Git in Rust with agents

Hi HN — we’re Julius, Jago, and Nils, and we’re building transload (transload.io).transload helps LTL trucking companies measure freight dimensions using the security cameras already installed in their terminals. Instead of sending shipments through a dedicated dimensioning station, we measure them automatically as they move through the normal dock workflow.We’ve put together a small HN-specific demo site here: <a href="https:&#x2F;&#x2F;hn.transload.io&#x2F;">https:&#x2F;&#x2F;hn.transload.io&#x2F;</a>In LTL trucking, dimensions matter because they affect pricing, freight classification, and trailer utilization. If a shipment is larger than the shipper reported, the carrier may undercharge for it while still giving up the same amount of trailer space. The obvious fix is to measure every shipment, but that is surprisingly hard in a busy freight terminal. Dedicated dimensioning systems work for freight that passes through them, but they can add forklift travel, create dock congestion, and change the normal flow of work. In practice, many terminals only measure a sample of their shipments.Jago grew up close to this industry through his family’s LTL trucking and cross-docking business. We did not start out building freight dimensioning. Our first idea was an AI system for optimizing forklift routes inside cross-dock terminals. After spending time with customers and talking to more than 50 trucking companies, we realized that forklift routing was not the pain people kept bringing up. Freight dimensions were.At the same time, we saw that spatial AI was advancing quickly. Monocular metric depth estimation has become dramatically better, making it possible to recover accurate 3D structure from ordinary camera footage without expensive LiDAR sensors. MapAnything (<a href="https:&#x2F;&#x2F;github.com&#x2F;facebookresearch&#x2F;map-anything" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;facebookresearch&#x2F;map-anything</a>) and MoGe (<a href="https:&#x2F;&#x2F;github.com&#x2F;microsoft&#x2F;moge" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;microsoft&#x2F;moge</a>) are two examples.Freight terminals also have helpful structure: fixed cameras, repeated workflows, barcode scan timestamps, and known layouts. Nearly every warehouse already has CCTV. That led us to a simple question: what if we could measure freight automatically using the existing security cameras, entirely in the background? That would allow carriers to measure every shipment without changing the dock workflow.Our system has two main steps: connect a barcode scan to the right object in the video, then estimate that object’s dimensions in real-world units.Dock workers already scan freight as part of the normal workflow. Each scan gives us a timestamp and a handling-unit ID. Around that timestamp, we analyze the video to infer which worker scanned and which shipment they scanned. We expected VLMs to handle this; they turned out to be far too unreliable. Instead, we train our own model that reasons in 3D over cues like gaze, body orientation, and movement.That association step is critical. A frame can contain dozens of pallets, several workers, forklifts, and partially hidden freight. If we attach the scan to the wrong object, the measurement is useless.Once we know the target shipment, we segment it and estimate a metric 3D bounding box from the monocular camera view. After the box is fitted, the dimensions are straightforward: length, width, height, and volume come directly from it.The hard part is precisely fitting that bounding box from one ordinary security camera. A single 2D image does not directly tell you object shape or scale, and many different 3D boxes can explain similar-looking image evidence. We use the object mask, visible edges, floor contact, camera geometry, and constraints from the terminal to find the 3D box that best matches the scene.We are currently working with several LTL carriers. For one customer, roughly 10% of checked shipments had dimension errors. The first use case is revenue recovery: identify under-dimensioned shipments, attach visual evidence, and help carriers correct the billing or classification. Longer term, the same data can help carriers understand trailer utilization better.LTL freight is an odd place to be doing 3D computer vision, and we learn something new every week. If you’ve worked on monocular reconstruction, 3D object detection, warehouse perception, or messy real-world CV, we’d love your take. Questions about freight, LTL terminals, or the technical approach are very welcome too.

Launch HN: Transload (YC P26) – Measuring freight items with CCTV

<a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2605.16487" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2605.16487</a>

A giant star may have destroyed itself in one of the rarest explosions

It's death

Let's Encrypt bans certificate usage in any US sanctioned territory [pdf]

Lies we tell ourselves about email addresses

Experience using AI software to prove Euler sum results [pdf]

Making Graphics Like it's 1993

Last April I shared about my Resonate project here (<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43694157">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=43694157</a>)A lot has happened since: the work I presented in much more detail at last June&#x27;s International Computer Music Conference (ICMC) got best paper award. I also gave a talk at the Audio Developer Conference in Bristol last November, the video is on YouTube).This year&#x27;s work, which I recently presented at this year&#x27;s ICMC, starts with known techniques from the phase vocoder literature to build self-tuning filter banks that extract very efficiently the frequency components that are actually present in the input signal. Overview on the project website, more details in the papers, including applications to super-resolution spectrograms and re-synthesis experiments.As many people have pointed out, none of the techniques I have used are new (some of them even have different names across different fields), but I haven&#x27;t seen them applied together in this way, and to me the results are incredibly satisfying and sometimes look magical. See for example this demo: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;LasdoIJJkw8" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;LasdoIJJkw8</a>Of course the best way to experience in person is through the free demo app: <a href="https:&#x2F;&#x2F;alexandrefrancois.org&#x2F;Oscillators" rel="nofollow">https:&#x2F;&#x2F;alexandrefrancois.org&#x2F;Oscillators</a>Looking forward to feedback from the community!

Show HN: Resonate – Low-latency, high-resolution spectral analysis

The LD_DEBUG environment variable (2012)

<a href="https:&#x2F;&#x2F;archive.is&#x2F;ZobUQ" rel="nofollow">https:&#x2F;&#x2F;archive.is&#x2F;ZobUQ</a>

FCC wants to kill burner phones by forcing telecoms to get all customers' IDs

Hi HN, I&#x27;ve been building Nucleus, a lightweight Linux container runtime focused on two workloads: ephemeral AI-agent sandboxes and declarative NixOS services. It&#x27;s a single Rust binary, no daemon.It is not a Docker replacement and not a strict subset of Docker either. I dropped the entire image-and-distribution half (no Dockerfile, no layers, no registry, no pull&#x2F;push, no persistent storage layer) in exchange for going deeper on isolation and reproducibility. The rootfs is either a directory copied into tmpfs (agent mode) or a Nix-built closure mounted read-only (production mode). If your mental model is &quot;run my image instead of docker run,&quot; this won&#x27;t fit. If it&#x27;s &quot;run untrusted or ephemeral workloads with stronger, auditable isolation on a single host,&quot; that&#x27;s the target.Things that I think are interesting:<pre><code> - Defense-in-depth defaults. All capabilities dropped, ~100-syscall seccomp allowlist (vs Docker&#x27;s ~300), up to 8 namespaces including time&#x2F;cgroup, Landlock LSM path ACLs per service.
 - Deny-by-default egress. Outbound traffic is denied unless you allow specific CIDRs or DNS-resolved domains. Enforced with namespace-local iptables rules.
 - Externalized, hash-pinned security policies. seccomp (JSON), capabilities (TOML), and Landlock (TOML) live as separate SHA-256-verified files, decoupled from the rootfs build. There&#x27;s a nucleus seccomp generate that records syscalls in trace mode and emits a minimal profile.
 - gVisor as a first-class integrated runtime, not an add-on. Explicit network modes including a gvisor-host mode that&#x27;s intentionally separate from native host networking.
 - Nix-native production path. nucleus.lib.mkRootfs builds locked-down closures; rootfs attestation verifies a per-file SHA-256 manifest at startup; first-class NixOS module.
 - Formal verification. TLA+ specs for the isolation&#x2F;resource&#x2F;filesystem&#x2F;security&#x2F;gVisor subsystems, checked with Apalache, plus property-based tests that drive the Rust implementation against the specs.
</code></pre>
Honest tradeoffs:
 - Linux x86_64 only. No macOS&#x2F;Windows&#x2F;BSD, no plans.
 - No CNI, no overlay networks, no cluster orchestration. nucleus compose is a single-host TOML DAG over systemd, not Swarm&#x2F;K8s.
 - Ephemeral-by-default storage. Persistence is opt-in via explicit --volume binds.
 - Agent mode applies several mechanisms best-effort by design (warn-and-continue on seccomp&#x2F;Landlock failure). For fail-closed isolation on ephemeral workloads use --service-mode strict-agent; for long-running services use production mode.Cold-start is ~12ms in the native runtime. Postgres 18 pgbench numbers under Nucleus are within noise of bare metal in our harness (full results in benches&#x2F;).