Edge is dropping uBlock Origin, leaving Firefox as the only major browser that still lets it fully block ads.
What the article says
- Firefox posted on Bluesky that its support for uBlock Origin isn't going anywhere.
- The announcement came right after Microsoft Edge said it's moving to the newer Manifest V3 extension system.
- That new system strips away the permissions ad blockers need to actually catch and stop ads and video ads.
- Once Edge makes the switch, Firefox becomes the last major browser where the full version of uBlock Origin still works.
What HN is saying
- Commenters note Firefox is unusual because Mozilla staff manually review uBlock Origin's code on every update to guard against smuggled malware.
- Several argue Google built Manifest V3 to look reasonable while conveniently gutting the one extension it can't compete with commercially.
- Others push back, saying uBlock Origin Lite still blocks nearly all ads for them and the outrage is overblown.
- A detailed reply lists what Lite actually loses: weaker tracker blocking, no CNAME uncloaking, and filter lists that only update when the extension itself does.
- A recurring thread wonders why people keep tolerating ad-choked browsing when a better, free option exists, with most blaming inertia and default browsers rather than ignorance.
Google built a compiler that lets AI crunch your data without ever seeing it, but it may be too slow to matter yet.
What the article says
- Google released HEIR, an open source compiler that converts normal AI models so they can run on encrypted data instead of plain data.
- The point is a server can process your information and hand back an encrypted answer without ever being able to read what you sent it.
- Google says the technique still costs a lot more computing power than normal AI, but that cost keeps falling.
- They showed it working on four things: content recommendations, credit card fraud detection, spotting suspicious network traffic, and detecting a wake word in audio.
- Hardware partners are building chips specifically to speed this up, and several universities are now doing research on top of the same toolkit.
What HN is saying
- Most commenters focus on the elephant in the room: real benchmarks suggest this is somewhere between a thousand and a million times slower than normal computing, and the post never states that plainly.
- A recurring counterpoint is that the simplest private AI already exists, just run an open model on your own hardware instead of encrypting anything.
- Some see a cynical angle, that this is really about letting Google keep scanning your data for signals like ad targeting or hotwords while claiming it never touched anything unencrypted.
- A cryptography researcher explains why some operations are far harder than others under this scheme, which is a big part of why it's so slow.
- Others push back, offering healthcare data sharing and cross bank transactions as places where regulation, not speed, is the real blocker today.
A comedian in a bin costume nearly beat Nigel Farage in a UK by-election, and it wasn't even close to a joke.
What the article says
- A comedian named Jon Harvey has run as a novelty candidate for years, first as Lord Buckethead against Theresa May, now as the bin headed Count Binface.
- In the Clacton by-election he came second to Nigel Farage, taking more than a quarter of the vote, his best result yet.
- Farage won comfortably, but Binface beat what past Conservative candidates used to get in that seat.
- UK election rules let anyone stand for a small deposit, which is why oddball candidates keep sharing the stage with prime ministers at the count.
- Binface has lost his deposit every single time but keeps getting invited back into the spotlight anyway.
What HN is saying
- Commenters swap Binface's actual policies, like taxing fossil fuel firms for solar panels, and find them more sensible than they expected from a joke candidate.
- Several point out the vote was really about Farage undeclaring a large crypto gift, and see Binface's strong showing as a rebuke rather than pure comedy.
- One detailed take argues the other parties skipped the race on purpose so Farage would win and stay as a weakened Reform leader rather than risk a stronger successor.
- A recurring worry is that joke parties aren't automatically the good guys, with an example from Hungary where one turned into personal enrichment.
- Someone compares it all to the Black Mirror episode where a cartoon bear nearly wins an election, noting reality got there first.
Maybe AI isn't out-thinking mathematicians, it's just never run out of scratch paper
What the article says
- The author argues AI's edge in math comes from a giant context window acting as scratch paper, not from deeper reasoning.
- Human working memory is tiny, so experts compress ideas into mental shortcuts, a trick called chunking.
- AI can hold hundreds of equations, dead ends, and assumptions in view at once without losing the thread.
- Math suits this trick because its symbols stay fixed and every step can be checked, unlike fuzzy real world questions like why a friend stopped texting back.
- The piece compares today's AI to physicist John von Neumann's speed and breadth, not yet to Einstein's deeper originality.
What HN is saying
- Commenters largely agree that brute force and endless patience explain a lot of AI's math success, since it never gets tired, bored, or discouraged.
- Several people say they're seen as smart despite having terrible memories, pushing back on the idea that memory and intelligence are the same thing.
- Others argue the opposite, that a strong memory for concepts really is what lets top mathematicians spot patterns others miss.
- A pointed thread claims the essay's author has a history of fringe race science writing, which colors how some read the piece.
- One coder notes that despite huge context windows, AI still misses nuance, like tracking who said what in an argument between two people.
A cryptographer argues AI is about to make software so bug free that spies and cops lose their favorite hacking tools.
What the article says
- A security researcher argues AI powered bug hunting is about to wipe out most of the hackable software flaws that police and spy agencies rely on to break into phones.
- For over a decade, agencies gave up asking companies for backdoors because they could just buy hacking tools to exploit bugs instead.
- Now that AI models are finding and patching bugs faster than ever, that supply of exploitable flaws is expected to dry up within a couple of years.
- He predicts this will push law enforcement to demand real, built in backdoors again, which is worse for everyone's security than the current setup.
- He admits he has no solution, just a warning that we are sliding toward a very different world.
What HN is saying
- A long thread on the history of wiretapping shows how surveillance went from expensive physical wires to something almost too cheap to meter.
- Commenters split on whether AI will actually reduce bugs, since some see companies shipping sloppier code faster even as AI fixes old flaws.
- Several people push back on the framing itself, noting cameras, phone metadata and tech company data already give agencies plenty of visibility, so calling it going dark feels overstated.
- One sharp reply notes any government mandated backdoor would be detectable by the same AI tools scanning for vulnerabilities, undercutting the secrecy backdoors need to work.
- A few commenters are simply unbothered by intelligence agencies losing hacking power, saying good riddance to mass phone surveillance.