Someone rooted his new Samsung TV and says it's quietly listening for more than you'd expect.
What the article says
- The author bought a 2025 Samsung TV and kept seeing oddly targeted ads.
- The firmware was encrypted and he couldn't get the keys, so he hunted for a software hole instead.
- He found a way to get a root shell on newer Samsung sets, which gives full control of the TV.
- Once inside, he says he found what he calls privacy horror. Merged threads claim the TV looks for microphones over Bluetooth and runs an image model on what the screen sees.
- The page itself is just a tweet teasing a write-up, so the details here are inferred from the thread and comments.
What HN is saying
- The top skeptic says this is old news: dog barking and baby crying detection are advertised features, and the finders let an AI convince them they'd uncovered a scandal. thread ↗
- Others push back that people learning by poking at their own devices is fine, and that you still have to ask why the TV needs to hear everything else to detect a crying baby. thread ↗
- A common thread is anger at ad-first design: no input button, a home screen full of ads, and settings that reset after updates. thread ↗
- Practical advice is nearly unanimous: never connect the TV to the internet, and plug in an Apple TV or Chromecast instead. thread ↗
- On the legal side, some want wiretapping prosecutions, but others point out that mandatory arbitration and the consent you click through make a lawsuit hard. thread ↗
Dan Luu says AI makes huge software projects cheap, so stop thinking small.
What the article says
- Luu backs a friend's claim that we should aim higher with software now that AI makes building so much faster.
- He built his own regex engine, a faster ripgrep fork and a quicker Rust build tool in a few hours, work that would once have taken a year.
- Skeptics dismiss rewrites like pgrust as slop, but he trusts it because its author tests hard and has found real Postgres bugs.
- Newer models make more things just work, yet he keeps moving to harder projects where they still struggle.
- An appendix from Yossi Kreinin warns most people should be less ambitious, since AI code can be messy and hard to verify.
What HN is saying
- Many commenters say AI lets them finally build long-shelved ideas, from a Kobo app platform to a JavaScript engine and a retro game remake. thread ↗
- A sharp pushback: a faster ripgrep or Postgres is hardly ambitious, and the real bottleneck is good ideas and solving other people's problems. thread ↗
- Quality doubts persist. Several people say AI code gets fragile and overfit, and demos rarely become products people use long term. thread ↗
- One thread argues team process, slow code review and unclear product direction now limit progress more than writing code does. thread ↗
- A thoughtful take compares this to factories first swapping one big motor for steam, and says the truly ambitious uses are not yet visible. thread ↗
- Some are weary of it all. One says other people's shared projects no longer excite them, and another doubts a 3 million line Postgres rewrite shows real skill. thread ↗
Big companies outside tech say AI has done nothing for their profits yet. Is that true?
What the article says
- Outside the tech giants, AI has produced no visible financial payoff. Revenue per worker and profit margins are rising for the big seven but flat for the other 493 companies in the S&P.
- Productivity growth across the economy still shows no lift from AI.
- Mainstream companies are starting to treat spending on AI as a cost that eats margins, not a route to growth.
- Cheaper models may not fix it. Business runs on judgment, and AI does not grasp accountability, trust, culture or fairness.
- The author's verdict: the AI story will be settled in the 493, not the big seven.
What HN is saying
- The most popular defence is patience. Electricity took a decade to pay off because factories had to be rebuilt around it, and AI may need new companies built around it. thread ↗
- A sharper reply says software is a special case. Compilers, tests and linters give coding agents a safety net that most other jobs simply do not have. thread ↗
- Some argue the gains exist but are captured by AI vendors. Others hit back that open models and competition would stop that, and the labs still lose money. thread ↗
- Many compare it to the 1987 line that computers show up everywhere but in the productivity numbers. Others note that this time the economy is propped up by huge debt. thread ↗
- Real-world stories split. One engineer says AI-found security bugs and pulled-off staff offset the coding speedup. Others describe office workers quietly cutting their workday to a few hours. thread ↗
- A common view is that the real blocker is people and management. Organisations change slowly, and bad management wastes any tool it is handed. thread ↗
A hacker left a Bitcoin prize on a PHP server to test the hype around a scary new exploit.
What the article says
- The author hid a Bitcoin key on a server running a PHP version from before the new PHP2Shell bugs were disclosed.
- First person to read it keeps the money, worth a few hundred dollars at the start and about two thousand once others chipped in.
- The page now says the prize was taken after just over 19 hours.
- A winner's writeup is linked, and the host promises a post about running the challenge.
What HN is saying
- Many saw the long wait as proof the panic was overblown. If exploits were this trivial, someone would have grabbed the cash within minutes. thread ↗
- Defenders of the hype said the setup was unfairly hardened, with a small request size cap and a stripped down server that breaks the exploit. Others pushed back that real sites vary, so each target takes real effort. thread ↗
- Some people claimed to hold a working exploit but refused to use it, fearing a honeypot. Critics called that convenient, and said the prize was too small to be worth revealing anything. thread ↗
- The sharpest turn came when the author announced a winner, who got in despite the tough config. One commenter said this shows it is exploitable within hours, even under harder conditions than most real sites. thread ↗
- People also tried AI models on it. Some refused on safety grounds, others went along with a gentle nudge, and the server mostly got knocked over by the traffic. thread ↗
A new systems language with no heap, no garbage collector and no borrow checker, still claiming to beat Rust and C++.
What the article says
- Goose is a memory-safe language where every growing value lives on a few compiler-managed stacks. Growing is a pointer bump, and leaving a scope frees everything at once.
- Nested data sits inline as one flat block with no pointers. A record that costs a heap allocation in C++ becomes a small packed run of bytes.
- References stay valid as arrays grow, and there are no lifetime annotations and no unsafe keyword. The compiler works out what each reference depends on.
- Threads share no memory at all, so data races can't exist. It compiles to C and the author's own benchmarks show it faster and leaner than idiomatic C++ and safe Rust.
- The catch: you have to plan where your data lives. The author says the repo was written almost entirely by Claude from his human design.
What HN is saying
- Many readers love the design but want the docs rewritten by hand. They can't tell which ideas are the author's and which are AI filler, and one commenter caught a wrong claim in the tutorial's first example. thread ↗
- The sharpest technical worry is how you handle many resizable things at once, like a hash table of changing strings. The author's answer: group data by lifetime, and use pools when deletion order is messy. thread ↗
- Some push back on the benchmarks, saying the C++ and Rust versions skip basic tricks like reserving capacity, so the comparison may flatter Goose. thread ↗
- A Rust user ported a few algorithms and saw big wins. Most of the gain came from the LLM picking flatter data structures, which carried back into Rust almost intact. thread ↗
- Others link it to older ideas: ML Kit's region inference, Apache request pools and Dart isolates. The author says Goose differs by leaning on huge 64-bit address space, so it's a poor fit for browsers and small chips. thread ↗