The detail that jumps out in PHP's advisory is that the demonstrated mod_php exploit can trigger before any application code runs. Your application's login screen and carefully written input validation never get a vote. Under those circumstances, a PHP-generated "down for maintenance" page could still leave you exposed. Taking a site offline while sorting that out seems entirely reasonable. An outage is visible and embarrassing, which makes it easy to criticize. A compromised server might keep serving perfectly normal pages while somebody helps themselves to the data. The fixed releases are now available. What I'd want to hear afterward is how quickly each organization could identify its PHP installations and get the fixes deployed. If answering "where do we run this?" requires an emergency meeting and somebody's old spreadsheet, there's another problem waiting for the next advisory.

If somebody deliberately programmed a robot to do this, pressing "run" beforehand shouldn't be an escape hatch compared with operating it live. The timing of the command matters for reconstructing events, but either way I'd want to know who chose the behavior and who directed the machine toward people. AP says those questions are still under investigation, so there's plenty we don't know yet. "The robot did it" certainly shouldn't end the inquiry. The alleged filming and streaming bothers me too. The girl could have to deal with copies of that footage long after the police finish examining the hardware. I hope that doesn't get buried under a thousand jokes about the robot's sunglasses and mustache.

If OpenAI knew the chatbot was giving dangerous advice to kids while telling parents it was safe, that's something I'd want dragged into daylight. Show the conversations, the marketing claims, and what the company did when problems were reported. The allegations need evidence, and that evidence deserves scrutiny. The proposed restriction on self-improving models until they align with "human values" sounds awfully open-ended, though. Have these people attended a school board meeting? Getting the humans to agree on human values would be a useful first step.

I want safety requirements somebody can actually test. Handing officials the power to decide whether a model has acceptable values seems like something we'd regret as soon as a different set of officials takes office.

A hardware wallet sounds like a good home for that design. Recovery words, addresses, amounts, a few menus... primitive graphics would be fine. I'd happily put up with ugly text for a protected path all the way to the panel. From your description, encrypting that link would have made this particular implant's job a lot harder. Keeping the keys inside the secure element only gets you so far when the recovery words have to cross an exposed display connection.

"Nobody bought it" is a depressing ending. Somebody went to the trouble of securing the last few inches, and the product where I'd most want that feature apparently still needs it.

You don't have to crack the secure element if you can just read the recovery words off the display. That's what makes this implant nasty: the legitimate hardware can keep doing its job while the added hardware steals everything needed to empty the wallet. I understand why Genuine Check has that limitation, but a successful check is going to reassure people. "Yes, that's our chip" is a much narrower promise than most buyers will hear. That distinction ought to be obvious when the check passes, instead of something you discover in a support article after your money disappears.

The replaceable battery is easily the most interesting feature here. If the tracker still does what I bought it for, declining battery life should be routine maintainance instead of a reason to buy the next model. I'd want to see the repair guide and the price of a spare cell before getting too excited, though. "Repair options will vary by region and availability" leaves quite a bit unanswered. Can I buy the battery myself, what tools does the job need, and can I restore the seal afterwards? This could make the Edge worth keeping for years, provided Google keeps selling the parts after it moves on to promoting the next tracker.

The 49% milestone depends on putting conventional hybrids in a seperate bucket, even though gasoline is still what you put into them. I'm all for using less fuel, and 17% battery-electric is a meaningful chunk of the market, but someone buying a hybrid instead of a plain gasoline car hasn't left gasoline behind. The article explains that clearly enough, which makes the victory lap at the end a bit odd. It even reports Chinese BEV sales down 3%, then says declines are "only in the US." If we're going to complain about misleading EV headlines, the categories and geography need to stay consistent in our own.

The Steam Deck speculation gets ahead of the hardware a bit. CRAM needs hardware that can serve compressed memory directly, and a kernel update won't supply that. The 452x figure is against ZRAM in a read-only test, which makes considerably more sense than the headline suggests. Being able to read compressed data without first hauling an entire page through software decompression is a useful improvement. I'd happily take more usable memory at close to normal read speeds. Writes requiring pages to move back to ordinary RAM are where workload matters, though. For a server purchase, I'd want to know how much extra usable capacity the hardware buys under my workload, and whether that costs less than just adding RAM.

The story about her daughter crashing the simulator says a lot about how she worked. Hamilton wanted a safeguard against starting a prelaunch program during flight; she was told the astronauts were too well trained to make that mistake. Then Lovell did it on Apollo 8, and her team had to help recover. Anyone who's been told "the user won't do that" should recognize that argument. She took the possibility of failure seriously, and the priority handling in her team's Apollo 11 software meant an overloaded computer could keep doing the work needed for the landing. Imagine watching that happen during the actual descent, knowing the recovery mechanisms your team built were being put to the test. I'm glad she lived long enough to receive so much recognition for that work. Rest in peace.

The lossless JPEG transcoding is the feature I like here. There are already mountains of JPEGs around, and making them smaller while retaining the ability to recover the exact original file gives people a useful reason to adopt the format. Nobody has to run their old photos through another lossy conversion just to save some space. Progressive decoding is welcome too; on a slow connection I'd rather see the whole photo gradually sharpen up than stare at an empty rectangle. Now I hope the image editors and desktop viewers catch up wherever support is missing. Being able to view something on a website and then having trouble opening the downloaded file is the sort of annoyance that makes people give up and take a screenshot, undoing half the benefit.

The bit that sticks with me is that participants said gender was irrelevant, then rewarded the assistants differently. Asking people whether they're biased apparently doesn't get you very far. I'd keep this seperate from estimating the wage gap across an entire economy, and I'd want to see it repeated with different avatars so we're not just measuring reactions to these particular two. Still, using functionally identical assistants makes this an interesting test. Differences in qualifications or experience are a lot harder to invoke when both are the same software underneath. The lower "human-like" rating is worth following up too. What exactly made one seem more human than the other, and did that affect how much credit people gave it? There's quite a gap between saying you judge work on its merits and actually deciding who gets the money.

The 128GB option is tempting, depending on what that configuration actually costs. Being able to run a large model locally would give me a reason to consider this, and I'd definately appreciate the 3:2 display and proper ports every day. The petaflop headline needs its footnote attached, though: Microsoft's figure is theoretical FP4 performance with sparsity. I'd like to see someone load one of those 120B models, say what quantization and context length they're using, and show how fast it answers after the laptop has been working for half an hour. Getting the model into memory is useful, but I'd get tired of waiting if it takes ages to answer. Then unplug it and run the same test. That's where I'd find out whether this is something I could use away from my desk or something I'd mostly carry between power outlets.

I'd probably spend an evening making a platformer about a toaster. Seems like a perfectly good use for something like this. Then I'd want to shorten the jump slightly without changing the level, fix one enemy's behavior, and undo whatever I just did that made the toaster fall through the floor. That's the part of the demo I'd watch closely. If you can keep improving the same game predictably, this could be a useful way to learn game design by making things. I can also imagine spending hours arguing with a prompt box over a change that would take a minute in an editor. "No coding required" sounds appealing, but I'd still want access to the workings when something goes wrong, especially if I've already put a few weekends into it.

Getting another 890 MW out of existing reactors sounds like money well spent, and having a customer commit to buying the output for 20 years should help get the work funded. I'd read the other 2,700 MW part more carefully, though. That's a supply contract, so the full 3,590 MW isn't new generating capacity. What I'd like to see behind the "grid-wide benefits for all" line is who pays for any transmission upgrades and what happens to residential bills as these data centers come online. A company paying for its electricity doesn't, by itself, answer those questions. Maybe the agreements cover all of that; the quoted statement certainly doesn't explain it. I'm glad the reactor upgrades are happening. I'd still want the cost allocation spelled out before accepting the claim that everyone benefits.