- 0 Posts
- 85 Comments
Many senior level “software engineers” are just tenured programmers and they’re managed by business people who don’t know software engineering either. One of the major benefits of using off the shelf software libraries is that they generally work as expected and have been through much more testing than something you just wrote, and often these libraries even receive free or cheap maintenance updates. You don’t want your developers wasting time reimplementing things and then wasting more time maintaining those reimplementations.
Getting the AI to write it is like mitigating the initial reimplementation cost by going to Fiver.
It’s not just anti-LGBTQ+. This is going to be bad for everyone. We’re just years away from banks and insurance companies factoring in your social credit score based on your activity on sites where you had to verify you’re not a legally considered a child.
You verify your age on Discord. Discord doxxes you through negligent handling of user data. Your account is found to be a member of a server that might suggest you are less responsible. Your rates are increased. Even if you know this is the reason, you cannot sue Discord because you were coerced into waiving your rights. The shareholders are happy because the line goes up. Is it already happening? I doubt any companies are rushing to tell us that they’re doing it, but the data is available to them.
They’re not that open because they steal Linux: https://web.archive.org/web/20220109040915/http://bbs.onyx-international.com/t/install-linux-or-alternate-os-and-gpl2-kernel-source/698
This is a real thing. It turns out more people than you would think are prone to paranoia and schizophrenia and if they spend too much time talking to LLMs they start to believe that the LLM is their friend, and then that the LLM is omniscient, and then from there it can spiral out of control as the LLM makes up a world of delusions for them. It’s hard to know how many people are affected because neither the victim nor the LLM realize there is a problem, and the problem is relatively new.
The start menu being React Native is irrelevant. If it were React in an Edge web view that would be a different story.
Opening the start menu should cause a spike in CPU usage. You want the CPU to open the menu ASAP instead of dragging out the process so the CPU usage is more flat.
But everything in Windows these days is wasting time stealing your data, loading ads or other unnecessary data from cloud services, and interacting with “AI.” Performance is one of the lowest priorities, somewhere between software quality and privacy. Since mid Windows 10, Microsoft consistently replaces things with modernized, but worse, versions and never returns to finish making the new version as good as the previous version that evolved over decades. It’s a really expensive way to ruin a product. They could make a React Native start menu where people wouldn’t complain about the performance. They probably did and people are only noticing now because of a recent regression.
That’s interesting. If you’re forking the Signal client to support archiving, you could also remove autoexpiration, the computer linking feature that is used when hijacking somebody’s account, and the ability to invite random people without clearance into your group chats. That would make it a reasonably secure and appropriate tool for communication over public infrastructure. We know they didn’t do at least one of these things.
3V at 100 microwatts significantly limits its usefulness.
They say they’re planning to make a 1W version, which I assume will be either be much larger or have a much shorter lifespan. How does it work? Does it have a way to stop the reaction or does the 1W battery generate 1W of heat when there’s no load attached?
The other problem is that the mouse does not click properly. Apple is still stubbornly refusing to put a second physical button in their mice. For almost 20 years they’ve been selling mice that can emulate right clicking by using a touch surface, but it seems like you still need to hold the mouse funny to avoid accidentally doing the wrong click because your other finger is resting on the other side of the mouse when clicking. At least they got rid of the little ball that likes to scroll horizontally while you’re scrolling vertically and gets clogged easily.
The author misses the irony of leaving Twitter, a for-profit, centralized, social network for Bluesky, a different, for-profit, centralized social network. Hopefully it’s different this time.
Children also learn to reading and writing using copyrighted works, often from borrowed books that they aren’t paying for. Some corporations would love if everyone had to pay individually, maybe per use, to access copyrighted material, and New York Times and American pro sport leagues would love if they could actually own recollections of copyrighted material, but neither of these is good for normal people.
https://www.eff.org/deeplinks/2023/04/how-we-think-about-copyright-and-ai-art-0
OpenAI is right. Almost everything of value on the internet is under copyright, and very little on the internet has clearly and unambiguously specified licensing information. If the software can only be trained on content that clearly allows training, the model isn’t going to “know” anything about anything since Steamboat Willie and it isn’t going to use broken dialects of older English from being limited to only public domain works that have been digitized and made available as public domain (reprints may not be public domain).
The article isn’t that clear, but the attacker cannot get Slack AI to leak private data via prompt injection directly. Instead, they tell it that the answer to a question is a fake error containing a link which contains the private data, and then when a user that can access the private data asks that question they get the fake error and clicking the link (or automatic unfurling?) causes the private data to be sent to the attacker.
They can most likely prevent further breakdown through software. If the meters and controls are functioning correctly, they can undervolt the CPU. But it’s not really a fix if that comes with a performance penalty. If it’s a bug where the CPU maxes out the voltage when idle so it can do nothing faster, that could be fixed with no performance penalty, but that seems unlikely.
Code pulled from GitHub or NPM can be audited and it behaves consistently after it has been copied. If the code has a high reputation and gets incorporated into bundles, the code in the bundles doesn’t change. If the project becomes malicious, only recently created bundles are affected. This code is pulled from polyfill.io every time somebody visits the page and recently polyfill.io has been hijacked to sometimes send malicious code instead. Websites that have been up for years can be affected by this.
I looked it up before posting. It’s illegal in 48 states, including California where most of these companies are headquartered, and every state where major cloud data centers are located. This makes it effectively illegal by state laws, which is the worst kind of illegal in the United States when operating a service at a national level because every state will have slightly different laws. No company is going to establish a system that allows users in the two remaining states to exchange revenge porn with each other except maybe a website established solely for that purpose. Certainly Snapchat would not.
I’ve noticed recently there are many reactionary laws to make illegal specific things that are already illegal or should already be illegal because of a more general law. We’d be much better off with a federal standardization of revenge porn laws than a federal law that specifically outlaws essentially the same thing but only when a specific technology is involved.
Web services and AI in general are completely different things. Web services that generate AI content want to avoid scandals so they’re constantly blocking things that may be in some situations inappropriate, to the point where those services are incapable of performing a great many legitimate tasks.
Somebody running their own image generator on their own computer using the same technology is limited only by their own morals. They can train the generator on content that public services would not, and they are not constrained by prompt or output filters.
What is RentAHuman’s cut? This is a very expensive service to operate. If an LLM posts a request for somebody to go pick up a package, what happens if the package never existed? What happens if the human just says that it never existed and takes the money or even the money and the package? Somebody in the middle needs to be arbitrating between AI agents that are notorious for making things up or getting details wrong and humans that just want to make quick money. Nobody is going to send requests if the humans are randomly stealing and nobody is going to fulfill requests if sometimes the request is unsatisfiable and you don’t get paid.