lily33 @ lily33 @lemm.ee

Posts

1

Comments

287

Joined

2 yr. ago

Since I'd firmly oppose such a war, I'm going to assume I was elected in a backlash against it.

So I'll just exit immediately, and instead make an agreement with another, non-invading, democratic country to help with rebuilding (using my country's money). I'll sign international treaties admitting the criminal nature of the invasion, pledge reparation, and agree never to do it again. Then start speed-signing disarmament treaties.

I assume I myself would get ousted at some point in this process though.

But wouldn't that just tell you, "Firefox was connecting to the internet"?

I have experience with GPT-4, and in particular I've used to for math questions in my work occasionally. I'm not sure how Bing chat compares.

For GTP-4, I've noticed the following:

1. How reliable the answer is depends on how easy or obscure the question is. It hasn't lied to me on easy or introductory material, but once your questions start becoming more obscure, and it's less likely to have the answer in the training set, it starts making things up.

• I think of it as search to an extent - it needs to have the answer in the training data to find it. Unlike google, it can usually find an answer even if you don't use the proper terms. But if it doesn't find an answer, it might make something up.
• "Easy or introductory" is relative - I have been able to get good answers for some masters-level math, and some wrong ones for lower-level things. Ultimately it depends on how much resources on the topic have been in the training set.

2. It's actually much more reliable in detecting errors than it's in generating text. So you can open a new chat and ask, "Is the following true: ..." and it will catch most of its own errors. Once it starts catching error, you should know you've left the reliable "easy questions" territory, and even if it can still be useful, exercise much more care.
3. The way you phrase a prompt matters a lot. For example, if you ask it to explain its reasoning step by step, it becomes much more accurate.
4. It is generally good in rephrasing questions to use better terminology.

.

Bing chat might be different in some regards. I know that it automatically searches the web for sources, and when generating an answer, and bases its answer on the contents of the sources it found - but I don't have experience with it.

That said, asking for additional sources (besides the search results it found) shouldn't improve the accuracy. It might just give you something you can use to fact-check it.

Earth. If there's any aliens in the Solar system, they can clearly hide from us pretty well, and I'm not really sure what they'd be interested in on any other planet...

What you're talking about is called "feature creep" and is a surefire road to poor quality.

I, for example, don't use any of the extensions you mentioned. And I checked two at random and both had less than 10k users, so they're by no means "must have". If they had to include all functionality that every "power user who does not appreciate having to frequently add new extensions" ever wanted, they might as well just rename it FireDinosaur or something. It will be both extremely heavy, and quickly extinct.

There is a difference between forefox-based browser and chromium-based one. Namely, if you base it on chromium, you take the blink engine and you can build watever UI around it you want. If you base it on firefox, you actually have to take the full firefox code and make changes to it.

All those firefox-based browsers are very similar to firefox with some small changes made. If you actually want to make large changes, keeping up with updates will quickly become a mess.

By contrast, qutebrowser has very little in common with Chromium except for the rendering engine - the user experience is totally different.

Also, bundling extensions with the browser is not the way to cater to power users - they will install the extensions they want anyway.

If gecko became embeddable (or better yet, servo was finished), so users could make alternative firefox-based browsers, that would be really good for power users. Right now things like qutebrowser are all based on blink, because that's the only option.

Almost any window manager should be able to do that. One way would be: timing WM + a script that opens each window in new workspace + bar configuration (if the built-in bar can't do what you want, there are plenty configurable thind-party bars that most WMs support).

That said, it should actually be possible to make a bullshit detector that detects bullshit writing.

Well, hypothetically, if someone defined the "consciousness" of every particle mathematically, and then figured out the laws that would allow us to compute (or at least approximate) the "consciousness" of a composite system (such as a brain), then we'd would have a genuine scientific theory.

I think calling it "dangerous" in quotes is a bit disingenuous - because there is real potential for danger in the future - but what this article seems to want is totally not the way to manage that.

I would say the risk of having AI be limited to the ruling elite is worse, though - because there wouldn't be everyone else's AI to counter them.

And if AI is limited to a few, those few WILL become the new ruling elite.

Since I don't think this analogy works, you shouldn't stop there, but actually explain how the world would look like if everyone had access to AI technology (advanced enough to be comparable to a nuke), vs how it would look like if only a small elite had access to it.

competition too intense

dangerous technology should not be open source

So, the actionable suggestions from this article are: reduce competition and ban open source.

I guess what it is really about, is using fear to make sure AI remains in the hands of a few...

As much as people here laugh - because yes, I get that it's very unlikely to work - I actually think this would be better for users than the ad-based model most social media use now.

Yes, but North Korea is something else. Cuba should be more representative of what the Soviet block was like.

I think the test for "free of copyrightable elements" is pretty simple - can you look at the new creation and recognize any copyrightable elements in it? The process by which it was created doesn't matter. Maybe I made this post entirely by copy-pasting phrases from other people, who knows (well, I didn't, only because it would be too much work), but it does not infringe either way...

From Wikipedia, "a derivative work is an expressive creation that includes major copyrightable elements of a first, previously created original work".

You can probably can the output of an LLM 'derived', in the same way that if I counted the number of 'Q's in Harry Potter the result derived from Rowling's work.

But it's not 'derivative'.

Technically it's possible for an LLM to output a derivative work if you prompt it to do so. But most of its outputs aren't.

One the contrary - the reason copyright is called that is because it started as the right to make copies. Since then it's been expanded to include more than just copies, such as distributing derivative works

But the act of distribution is key. If I wanted to, I could write whatever derivative works in my personal diary.

I also have the right to count the number of occurrences of the letter 'Q' in Harry Potter workout Rowling's permission. This I can also post my count online for other lovers of 'Q', because it's not derivative (it is 'derived', but 'derivative' is different - according to Wikipedia it means 'includes major copyrightable elements').

Or do more complex statistical analysis.

Indeed not a conspiracy. A conspiracy involves a crew of conspirators conspiring to do something illegal or unethical.

If it's just the laws of physics conspiring, it doesn't count.