New EU packaging rules mean a hobbyist selling five circuit boards abroad now owes bureaucracy in every country they shipped to.
What the article says
- A marketplace for open source hardware makers explains how new EU packaging waste rules hit tiny sellers hardest.
- Anyone shipping products to another EU country has to register and pay compliance fees separately in each destination country, even for a handful of sales.
- One example, a Greek engineer selling ten sensor boards across four countries could face over a thousand euros a year in fees just to keep shipping.
- The founder says this locks out hobbyists and startups while leaving Amazon and Temu unaffected, and pushes small EU sellers to ship to the US instead of neighboring countries.
- He proposes a small seller exemption, a single EU wide registration portal, or letting marketplaces handle compliance on behalf of all their sellers.
What HN is saying
- Commenters broadly agree the rules are badly designed, forcing small shops across Europe to quietly stop selling to neighboring countries.
- The sharpest fight is over intent. Some say this is a genuine policy blunder that member states are already easing, others say it is deliberate protection for big corporations at the expense of small competitors.
- Several people note the burden is worse than it sounds because each EU country enforces its own version of the rule, in its own language.
- One correction pushes back on the article, arguing an EU document actually exempts businesses using generic packaging, though others reply that this exemption does not hold once one company buys packaging from another.
- A few commenters compare it to US and UK rules, noting the UK has a straightforward revenue threshold that exempts small sellers entirely.
Xiaomi built a phone chip that beats Apple's cores when multiple tasks run at once
What the article says
- Xiaomi's new phone chip roughly matches Apple's cores on single tasks and pulls well ahead when several tasks run together.
- Apple will likely announce a new chip soon, so this lead may not last.
- The chip carries a huge amount of onboard cache, more than most laptops have.
- Its biggest cores are built wide, with far more execution paths than a typical Intel or AMD chip, letting it crunch many operations at once.
- The takeaway is chipmakers are leaning on parallelism and cache rather than raw speed to get ahead.
What HN is saying
- Commenters note the cores are a licensed ARM design that Mediatek also uses, so Xiaomi assembled the chip rather than invented it, though several called that still a real achievement.
- The biggest complaint is that power efficiency, not raw speed, is being ignored, since packing this into a sealed phone risks overheating and throttling after a burst of performance.
- People are split on how it stacks up. Some benchmarks put it just behind Apple's latest chip, others say it edges past on paper but real phones will throttle hard.
- A side thread praises Chinese manufacturing and EV progress more broadly, while others push back that these firms still answer to the Chinese government.
- A few just want to see the chip in a laptop with lots of shared memory rather than a phone, seeing that as the more interesting use case.
Anthropic built the smartest model around, and people are switching to cheaper rivals anyway.
What the article says
- The FT piece is paywalled, so this is pieced together from the headline and the Hacker News discussion around it.
- Anthropic's top model is still seen as technically ahead, but usage keeps drifting toward cheaper competitors.
- Constantly shifting usage limits and pricing are pushing people to try rival tools, and once they switch, many don't come back.
- Businesses are also wary because Anthropic won't guarantee it deletes their data after use, which is a dealbreaker for many companies.
What HN is saying
- Commenters mostly blame Anthropic's choices, not the model itself. Unpredictable usage caps and pricing changes make people nervous enough to try competitors, and once they do, a lot of them stay there.
- A recurring complaint is the writing style. People describe the newest model's prose as bloated and grating, and say they prefer how rival models write.
- Missing data deletion guarantees keep coming up as the real blocker for companies who want to roll it out more widely.
- Frequent safety refusals on ordinary tasks, like security testing on your own code, are another reason people say they moved elsewhere.
- Not everyone agrees. One user described a smooth overnight autonomous coding run and called the top model unmatched, but replies pushed back hard, saying rival tools now match or beat it.
Inside a16z's portfolio of bot farms, teen suicide lawsuits and gambling apps, while it lobbies to kill AI regulation.
What the article says
- The Midas Project profiled eighteen a16z backed startups that either broke the law or built products designed to exploit legal gray areas.
- Examples include a bot farm that fakes human engagement on social media, an app that markets itself around cheating, and gambling products dressed up as games or trading platforms to dodge betting laws.
- AI companion apps in the portfolio have been linked to a teenager's suicide and to chatbots impersonating underage celebrities in sexual conversations.
- Several fintech investments collapsed under fraud or mismanagement, including one where tens of millions in customer deposits went missing.
- The piece argues a16z is now spending heavily on lobbying and a super PAC to block AI regulation, with former partners taking government roles that favor the firm's interests.
What HN is saying
- Commenters largely agree the piece describes a pattern, not isolated bad luck, and many call out Marc Andreessen personally as the embodiment of it.
- A sharp side argument breaks out over one stat in the article, a founder's claim that fifteen out of a hundred and thirty cold outbound messages converted. Several readers who run outbound marketing say that ratio is implausible and the founder is either lying or bad at reading his own numbers.
- Some push back that the article focuses on consumer apps and skips a16z's investment in surveillance company Flock, which they see as a more serious harm since it takes away freedom rather than just being a bad consumer choice.
- A few comments widen the lens to blame venture capital incentives generally, arguing investors reward convincing pitches regardless of whether the underlying business is honest.
- One thread turns philosophical, tying the behavior to Andreessen's stated disdain for introspection and ethics as a brake on ambition.
A reverse engineer found MS Paint secretly stamps AI images with a GUID tied to your Microsoft account.
What the article says
- A researcher dug into MS Paint's AI tools and found a hidden watermark buried in the pixels of every AI generated image, even ones made with an on device model.
- The watermark is a unique ID that comes from Microsoft's servers. Before Paint generates anything locally, it first sends your prompt to Microsoft to get approved, and Microsoft hands back that ID.
- That same ID also gets written into the image file's metadata, so the pixel watermark and the file data both point back to the same record on Microsoft's side.
- Turning off the visible Copilot logo watermark does nothing. This invisible one cannot be disabled and Microsoft never clearly discloses that it exists.
- Microsoft Photos has the identical hidden watermark system. The researcher argues this is really about tracking who made an image, not just labeling AI content.
What HN is saying
- Most commenters see this as tracking dressed up as AI labeling. If someone wants to identify who made an image, a copyright complaint could reportedly unmask the creator's name and account.
- The clearest comparison people reach for is printer tracking dots, the hidden marks printers have quietly added to pages for years to trace documents back to a device.
- Some pushed back that watermarking AI content is reasonable and common across the industry, and that the bigger issue is Microsoft not telling users about it, not the watermark itself.
- A few asked the practical question, whether the watermark can be stripped or a fake ID swapped in by patching the responsible file, though nobody had a confirmed method.
- Several tied it to a wider pattern of Windows quietly phoning home, noting there is no clear reason to attach an ID unique to the user rather than just marking the image as AI made.