File preview handlers can be written by third parties so this is impossible without writing a completely new system.

Preview should work again once the user opens the file and confirms the action in the security dialog informing them it is from an untrusted location. This is just bringing preview functionality inline with file open functionality to treat the file as untrusted.

Computers are very reliable at reporting memory usage. The problem is memory usage is more than a single number, it is complex. Do you want memory allocated or memory used? Do you want to include memory shared with other processes (which should only be counted once system wide) or not? Do you want to include memory swapped to disk or not?

Just another piece of data websites can use to fingerprint you with.

Don't forget the TCP evil bit!

The whole point of LLMs is they will recognize and reproduce patterns of words. If you have control over the patterns they are trained on, you can control what they reproduce. There's nothing particularly surprising to me here, this is how I would expect it to work.

Pandora's box can't be closed. AI is here to stay.

It is impractical to sue all the users, so regardless of who should be responsible, they will try to sue the people running the AI. However, it will soon become impractical to sue all the people running AIs as well, if it's not already. Pandora's box is opened, the genie is out of the bottle, the horse has left the barn, etc.

I was watching a video that pointed out the devices these SIM cards were in could only use a small number of the cards simultaneously. And even if you ignore this, 100,000 sim cards is a small fraction of the total number of legitimate devices in NYC, so simultaneous usage would not have had much effect. So they are probably just used for mass-spamming and similar campaigns.

Which means is it going to quickly fall out of date for any "advice" it gives. It is also not intended to give stock advice so it will be naturally bad at it and more likely to hallucinate advice.

I just had a meeting today where we discussed an AI project I'm working on and apparently some in the US military are nervous about letting AI into classified areas, because they can't peek under the hood and see how it works. To be fair I think this is something of a requirement for such software in general. Specific concerns are things like, how do they know it won't go rouge thanks to training data from a foreign adversary? AI doesn't work like that. It generates words (or other pieces of information like images or audio or video) that can potentially deceive, and it can make use of tools that the developer explicitly provides only in the ways the developer allows them to be used. That's it. So I would worry more about the competency of the developer since if an AI is able to do damage it says more about the developer than the AI, I feel.

Last i tried it I had three problems with it. #1 was it is not intuitively located and it's easier to Google how to cancel it to jump right to the page than it is to navigate Amazon's own site to find it. #2 was it was unclear that your subscription will remain active until the end of the billing cycle until after you've cancelled, which may encourage customers to delay cancelling which increases the odds they will forget and get billed for another month. #2 was there was an excessive number of "please don't cancel, you'll lose X" prompts each of which was formatted differently to make cancelling more difficult.

It's perfectly reasonable a new OS version has higher system requirements. It's just in this case MS is pushing them to ensure manufacturers create PCs that can support certain security features. For example I understand TPM can help enforce boot security and disk encryption key storage. Good stuff to keep secure.

But if they want to throw out their PC and get a new one instead of hardware upgrading (if possible, not always), or even just switching OSs entirely for Linux (free, the only cost is time to learn), that is their decision, but they HAVE a choice, even if they don't want to admit it, or don't want to even research to understand the choice they have.

If you're taking exams in a web browser you can already just plug the questions into Google's website and get Gemini to answer them there. This is just a convenience feature and doesn't change what students actually have access to. Even before AI, if you have access to Google (required for Gemini) nothing is stopping you from doing normal web searching for answers. Nothing is new here, except educators getting cold water splashed on their face as to how technology has been advancing when they weren't paying attention.

Someone was using those pictures as placeholders as they were not provided with actual camera pictures, and they either forgot to replace them or the real pictures were just added to the mix and the stock photos remained. This is why placeholders have to be explicitly marked as such or otherwise be something nobody would sign off on (I used nyan cat for such things myself).