HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-08-11

Meta Muse Glimmer – open weights 30B local coding model (Meta AI)

Meta open-sources Muse Glimmer, a 30B-parameter model for local agent workflows.

1142 pts · 616 comments

Docker Sandboxes – Disposable, isolated sandboxes for AI agents (Docker)

Docker launches disposable microVM sandboxes for running AI coding agents unattended.

Login requirement frustrates users; many share open-source alternatives.

666 pts · 370 comments

France to ban unsolicited telemarketing calls (lemonde.fr)

France will ban unsolicited telemarketing calls starting August 11, 2026.

Widespread support, with skepticism about enforcement against scammers and impact on Moroccan call-center jobs.

594 pts · 326 comments

Tl;dv: Over 180k meetings left wide open (bobdahacker.com)

A security researcher found tl;dv's Firestore database exposed 181,874 meetings to any authenticated user.

Commenters expressed outrage at the CEO's six-month inaction and criticized Firebase's insecure defaults.

592 pts · 193 comments

The UK's war on anonymity has come to America (Effort)

UK NGOs use child safety rhetoric to push digital ID laws in US states.

Thread splits: child safety as pretext vs. genuine parental concern driving regulation.

569 pts · 602 comments

Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models (ft.com)

Zuckerberg criticizes closed AI rivals as Meta returns to open models.

Skepticism dominates, with many seeing open-sourcing as a losing player's strategy to change the rules.

555 pts · 525 comments

As AI eats the web, the internet’s collective memory is disappearing (thewalrus.ca)

AI summaries and platform decay erode the internet's archival function.

Skepticism toward the article's premise, with many citing Google's decline but rejecting the anecdote as weak.

521 pts · 608 comments

Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots (Cactus Compute)

Cactus Compute releases Needle 2, a 45M-parameter LLM in a 14MB binary for tool calling on sub-$200 devices.

The demo showed false positives, but the team cites the built-in confidence score as the intended mitigant.

419 pts · 154 comments

What Happened to HackerOne? (Curiosity With a Side of Chaos)

HackerOne's shift from hacker focus to VC-driven profits degrades platform and community.

VC money and sales focus corrupted HackerOne's hacker-centric mission and community.

377 pts · 198 comments

Mars Bar from 1991 found – and it's 20g bigger than today's (bbc.com)

A Mars Bar from 1991 found in a house clearance is 56% bigger than today's version.

Thread debates shrinkflation's greed vs. health effects, with a joke about Mars's marketing motives.

373 pts · 562 comments

H3-metal – Native MiniMax-H3 inference for Apple Silicon (GitHub)

antirez's h3-metal runs MiniMax-H3 video inference natively on Apple Silicon with Metal.

Most users report the model works on 64 GB Macs, but speed is far slower than on Nvidia GPUs.

354 pts · 77 comments

Illinois just passed a law that puts Linux on the hook for age verification (Linux Stans)

Illinois HB5511 requires OS providers to build age verification by 2028, with no open-source exemption.

Two camps: one says enforcement is impossible for hobbyists, another warns Red Hat/IBM will comply by network effects.

328 pts · 465 comments

Auto mode is now the default in Claude Code (Claude)

Auto mode, which routes tool calls through a safety classifier, becomes the default in Claude Code.

Pro-sandbox sentiment dominates: many run auto-approve in containers and reject classifier-dependent safety.

283 pts · 308 comments

Squeak 6.1 (Squeak.org)

Squeak 6.1 'Vanessa' releases after 4 years with 1700+ patches.

Smalltalk's influence on JavaScript and the persistent image model dominate the thread.

275 pts · 135 comments

How Claude marks AI-generated content (Claude Help Center)

Anthropic details how Claude embeds imperceptible watermarks in AI-generated text worldwide.

Technical debate on watermarking methods, with skepticism about detectability and circumvention.

273 pts · 245 comments

Show HN: Scroll through all 43252003274489856000 Rubik's Cube states (Every Cube)

EveryCube lets you scroll through all 43 quintillion Rubik's Cube states.

The thread is lighthearted, appreciating the absurdity of doom-scrolling through 43 quintillion states.

239 pts · 82 comments

Chicken Scheme 6.0 (code.call-cc.org)

Article body wasn't reachable. HN discussion still summarized.

Users praise the new Unicode support and debate Scheme vs. Common Lisp for serious projects.

238 pts · 32 comments

Learning more about Claude's mathematical capabilities (anthropic.com)

Claude increased the proven lower bound of Riemann zeta zeros on the critical line from 41.6% to 67.2%.

Mixed reactions: genuine mathematical progress versus frustration with Anthropic's anthropomorphizing language.

233 pts · 149 comments

Stop Killing Games: It's time to sue Sony, join us (massaschadeconsument.nl)

Dutch foundation files class action against Sony over high PlayStation Store game prices.

Debate: Sony's monopoly claim vs consumers' freedom to choose other platforms.

231 pts · 128 comments

Parametron: 50s Japanese computer that uses neither transistors nor vacuum tubes (ethw.org)

Eiichi Goto invented the parametron in 1954, a cheap ferrite-core logic element used in Japan's early computers.

Parametrons were a dead-end technology due to speed and scaling limits versus transistors.

227 pts · 56 comments