Meta just gave away a free coding assistant good enough to run entirely on your own laptop.
What the article says
- Meta released a new model called Muse Glimmer, and unlike most of its rivals, you can download the whole thing and run it yourself.
- It is built to run on a single gaming graphics card, no cloud connection needed, and still handle real coding and agent tasks.
- Meta shrank it using compression tricks so it fits in less memory, plus a trick that lets it write several words at once instead of one at a time, so it feels faster to use.
- It can also look at screenshots and images, not just text, and it understands over a hundred languages.
- Meta is giving it away under a very permissive license, so anyone can use or modify it freely.
What HN is saying
- Commenters mostly compared it against Qwen, a rival open model, and opinions were split on whether Glimmer actually wins or just arrived first to avoid losing that fight.
- Several people are already running it on ordinary home hardware, from Mac minis to gaming GPUs, and reporting it works, just slowly.
- A recurring theme was skepticism toward Meta as a company even while praising the release, with one person calling it a marketing move dressed as generosity.
- A few users hit real problems, including one who said it got stuck in a confused loop trying to debug some broken code.
- There is genuine excitement that local, no-internet AI assistants might finally be practical rather than just a geeky experiment.
France is banning cold sales calls entirely, and the fines are big enough that people think it might actually work.
What the article says
- France bans unsolicited telemarketing calls starting August 11, requiring businesses to get consent first, and consent can be withdrawn anytime.
- Most French people get at least one unwanted sales call a week, and consumer groups pushed hard for this after years of complaints.
- Fines run steep, up to a few hundred thousand euros per illegal call for companies, with a website for reporting violators.
- Exceptions remain for calls you've opted into or companies you already do business with.
- The law worries Morocco, where tens of thousands of call center jobs depend heavily on the French market.
What HN is saying
- Commenters agree the real menace isn't legitimate telemarketers but scammers who ignore any list or law anyway.
- Several people say the actual fix is technical, blocking spoofed caller IDs, since scam calls often fake a local number to seem trustworthy.
- Readers from Norway, Finland, and Spain describe opt out systems that genuinely cut spam once enforced, contradicting the pessimists.
- A popular French app called Saracroche gets repeated praise for filtering junk calls better than any law so far.
- One sharp aside notes the government's own do-not-call list, Bloctel, recently suffered a data leak of its own.
Google search is quietly forgetting things, and AI is speeding up the rot in the web's memory.
What the article says
- Google's AI summaries are now getting basic facts wrong, like inventing sunset times, which points to a deeper problem than sloppy answers.
- Search used to be a reliable path to a stable body of knowledge. Now pages vanish, archives get sued or blocked, and AI answers replace clicks so nobody notices the sources disappearing.
- Wikipedia, the Internet Archive, and sites like FiveThirtyEight are all losing traffic or funding because AI tools scrape and summarize their content instead of sending readers there.
- The author argues search and archiving should be treated as public infrastructure, pointing to countries building government backed alternatives to Google.
What HN is saying
- Many commenters say they already left Google for DuckDuckGo or the paid search engine Kagi, and don't miss it.
- A recurring theme is that keyword search skills are dying out, so people can no longer find niche or older material that AI chatbots simply don't surface.
- Several people shared stories of AI assistants confidently giving wrong or dangerous advice, from car repairs to VPN setup, while others had genuinely great experiences, so opinions split hard on how useful AI search really is.
- One sharp exchange debates whether the Internet Archive brought its legal trouble on itself by ignoring writers' groups for years, versus the ruling itself being an unjust reading of copyright law.
- A few commenters push back on the article's framing, arguing the web's decline started well before AI, driven by SEO, walled gardens, and ad driven content.
Docker launched throwaway virtual machines for AI coding agents, and the comments are a goldmine of rival open source tools.
What the article says
- Docker Sandboxes give each AI coding agent its own disposable microVM, not just a container, so it can install packages, run its own Docker containers, and get destroyed when done.
- The pitch is letting agents run in full autonomy mode without approval prompts, since the microVM keeps mistakes off your actual machine.
- Only your project folder gets mounted in, and you set the network and filesystem rules yourself.
- It works with most major coding agents including Claude Code, Codex, and Copilot CLI, though right now only on Mac and Windows.
What HN is saying
- A Docker engineer showed up to clarify these are real lightweight virtual machines with their own kernel, built on a new custom hypervisor rather than containers or Firecracker.
- People are annoyed the tool requires logging in, which breaks scripts that spin sandboxes up and down automatically.
- The thread turned into a swap meet of alternatives people built themselves, with names like yoloAI, Locki, smolvm, and Agent Circus all offering similar isolation without a login wall.
- One sharp reply pushed back on the whole premise, arguing sandboxing is a patch rather than a fix and that real safety needs proper tiered permissions, not just walling agents off in a box.
- A few commenters noted the isolation strategy matters because recent AI agent security incidents were not actual sandbox escapes but misconfigured rules the agent was technically allowed to use.
British child safety groups are quietly pushing the same digital ID laws that now jail UK dissidents onto American states
What the article says
- An investigation traces five British nonprofits lobbying US states to pass digital ID laws, all pitched under the banner of child safety.
- In Britain those same laws now help track and arrest political dissidents, and one group behind them is pushing to ban VPNs too.
- The groups overlap heavily. Shared board members, shared funding, and lobbying that went undisclosed for over a year in some cases.
- One group, 5Rights, was founded by a member of the UK's House of Lords and helped design a California bill modeled directly on British law.
- Millions in government contracts flow to these groups, with internal reports openly describing a goal of controlling online speech, not just protecting kids.
What HN is saying
- Commenters push back on the framing, noting US states passed age verification porn laws before Britain did anything comparable.
- A popular fix floated in the thread, a phone level child mode with a government whitelist, so ordinary sites never need ID checks at all. Others worry that just invites locked down hardware instead.
- A recurring theme is that this is really about big tech dodging liability, with child safety used as cover by both sincere parents and cynical operators.
- One widely discussed comment argues surveillance is now cheaper than privacy, and the bigger risk is rogue cops or abusive exes with access to these tools, not the NSA.
- Some accuse the bill's sponsor of being a naive tool of dark money, while others think she knows exactly what she's doing.