Can you give examples of things that were called genocide but actually weren’t and just watered the word down?

Yeah, it’s good enough that it even had me fooled, despite all my “it just correlates words” comments. It was getting to the desired result, so I was starting to think that the framework around the agentic coding AIs was able to give it enough useful context to make the correlations useful, even if it wasn’t really thinking.

But it’s really just a bunch of duct tape slapped over cracks in a leaky tank they want to put more water in. While it’s impressive how far it has come, the fundamental issues will always be there because it’s still accurate to call LLMs massive text predictors.

The people who believe LLMs have achieved AGI are either just lying to try to prolong the bubble in the hopes of actually getting it to the singularity before it pops or are revealing their own lack of expertise because they either haven’t noticed the fundamental issues or think they are minor things that can be solved because any instance can be patched.

But a) they can only be patched by people who know the correction (so the patches won’t happen in the bleeding edge until humans solve the problem they wanted AI to solve), and b) it will require an infinite number of these patches even to just cover all permutations of everything we do know.

Here’s an example I ran into, since work wants us to use AI to produce work stuff, whatever, they get to deal with the result.

But I had asked it to add some debug code to verify that a process was working by saving the in memory result of that process to a file, so I could ensure the next step was even possible to do based on the output of the first step (because the second step was failing). Get the file output and it looks fine, other than missing some whitespace, but that’s ok.

And then while debugging, it says the issue is the data for step 1 isn’t being passed to the function the calls if all. Wait, how can this be, the file looks fine? Oh when it added the debug code, it added a new code path that just calls the step 1 code (properly). Which does work for verifying step 1 on its own but not for verifying the actual code path.

The code for this task is full of examples like that, almost as if it is intelligent but it’s using the genie model of being helpful where it tries to technically follow directions while subverting expectations anywhere it isn’t specified.

Thinking about my overall task, I’m not sure using AI has saved time. It produces code that looks more like final code, but adds a lot of subtle unexpected issues on the way.

Nice, enjoy!

An alternative that will avoid the user agent trick is to curl | cat, which just prints the result of the first command to the console. curl >> filename.sh will write it to a script file that you can review and then mark executable and run if you deem it safe, which is safer than doing a curl | cat followed by a curl | bash (because it’s still possible for the 2nd curl to return a different set of commands).

You can control the user agent with curl and spoof a browser’s user agent for one fetch, then a second fetch using the normal curl user agent and compare the results to detect malicious urls in an automated way.

A command line analyzer tool would be nice for people who aren’t as familiar with the commands (and to defeat obfuscation) and arguments, though I believe the problem is NP, so it won’t likely ever be completely foolproof. Though maybe it can be if it is run in a sandbox to see what it does instead of just analyzed.

Oh I agree, but as long as they can find people to buy their weapons (even on credit that depends on successfully using them) or weild them for them, they won’t be going anywhere. And automation is quickly approaching the point where they won’t even need others to choose to fight.

Problem with arms races is you can’t end them unless everyone agrees to end them, otherwise you just forfeit.

That “refusing to continue communication” might have even just been “couldn’t hear or feel vibrations from incoming calls”. It’s also possible he thought they weren’t being helpful and decided it was a waste of time to rely on them (all depends on how that initial call went, though the fact that they say he didn’t ask for help but he says he did could suggest a communication breakdown or tone mismatch).

It did sound like he was unprepared for how to handle such an emergency if they didn’t even use the warming gear they had. But the question is at what point does unpreparedness become criminal and did he really have extra responsibility for her safety even if he thought they were equally experienced, or that she was at least experienced enough to handle her own safety? Unless the defense is lying completely, it sounds like the prosecution isn’t approaching this in good faith and might be seeking revenge instead of justice.

What a fucking whiny loser. Gets caught cheating and starts crying about how curling is based on trust probably because he exploited that trust to get to the olympics in the first place.

Send him home. Even if it costs the team any medals they have a chance at, better to shut that shit down hard than stand by it to get medals that will be tainted by the whole thing anyways.

Glad I already dgaf about curling or the olympics, otherwise I’d be concerned that remarks like that might make people think that curling should just be a casual backyard sport that doesn’t belong in the Olympics if “trusting your opponents” is more important than “following the rules”.

What a dumb fucker. Hope he doesn’t have a lot of other trash like him to rally around his worthless take.

It’s two different things being argued about: the legal term “hacking” vs the every day language term, which I believe implies something more specific than “unauthorized access”, something where technical or social skills were used to gain that access.

That’s the parallel I was trying to draw by mentioning the word “hotwiring” instead of “stealing”. It would be like if the legal term for stealing a car was “hotwiring”.

That said, I did see that the OP of this tangent is actually trying to argue the “this isn’t illegal” angle rather than the difference between legal terms and broader language terms.

I agree this falls under the legal definition of hacking, but I also agree with those basically saying that this falls outside of the way they think the term should be used. It waters down its meaning.

Yeah but if I did, no one would say I hotwired the car.

I switched about a year ago to fedora cinnamon. Less frustration than windows, even though cinnamon kinda sucks compared to KDE that I switched to immediately after the first time I tried it (should have tried it months sooner, literally only took a few mins to install and check out).

While I wouldn’t say that there were zero problems, I did notice that I spend less time troubleshooting or searching for how to change something on Linux than I did on windows by the end. Also, going from empty disk to gaming involved fewer steps, at least with an AMD gpu.

Not sure if you’re exaggerating the low resolution, but I haven’t noticed quality issues on Amazon. I doubt the stream I’m getting is 4k, but it’s certainly better than 720p.

I’m using the flatpak firefox from the fedora install instructions that comes with more codecs, though. It plays a bunch of video that VLC won’t render with my current setup and I haven’t yet put the effort into getting full codecs outside of Firefox yet, but maybe your system has a similar codec situation and prime video defaults to some old or neglected format that caps out at the res you see.

Or it could be what you think and for some reason my system isn’t triggering it. Argh, this future is annoying.

Seems like this is possible, but the method (and maybe ability) depends on your window manager.

If you’re using x11, you can interact with the window manager via the command line (so could set up the whole thing in a script). An example command line tool: xdotool (search for “interacting with x11 via command line” for more info).

If you’re on Wayland, one of the design principles was to avoid programmatically interacting with window size or position; the user will set up their composer to behave as they want, not how the programmer of that program wants, and especially not how programmers of other arbitrary programs want (it was a security issue at the extremes, or could be annoying for more common cases). But you, the user, do have control, though it depends on your DE and what Wayland compositor it is using. On fedora KDE, you can use KWin scripts (which supports several languages).

There’s also some other window managers that can offer better control, and perhaps it’s enough for the window manager to simply remember the position of windows when they are closed (which I think Wayland does or can be configured to do easier than writing a script, then you just need a launch script for the programs in your shortcuts).

Yeah, I wasn’t sure whether mine wasn’t working due to hardware issues or the software. Haven’t dug in to debug it much because I don’t really use it, but the times I have tried, it failed. It could sometimes connect to some devices but the connection wouldn’t be stable and no useful data transferred before it dropped.

Not even sure I have the tools to debug it. Does BT have a loopback mode to test if bits in are the same as bits out? Does wireshark work at the physical link level?

They might have set up the user agreement for it. Stackexchange did and their whole business model was about catching businesses where some worker copy/pasted code from a stackexchange answer and getting a settlement out of it.

I agree with you in principle (hell, I’d even take it further and think only trademarks should be protected, other than maybe a short period for copyright and patent protection, like a few years), but the legal system might disagree.

Edit: I’d also make trademarks non-transferrable and apply to individuals rather than corporations, so they can go back to representing quality rather than business decisions. Especially when some new entity that never had any relation to the original trademark user just throws some money at them or their estate to buy the trust associated with the trademark.

Yeah, when you’re compiling from source code, it’s much easier to add in ARM support. Kernels do need to handle more hardware-specific stuff, like interrupt handling, context switching, feature enablement and the like will likely need custom ASM code and might have different parameters/events to handle. But at the user level, you can often compile for ARM by just changing a few command line arguments and it’ll be fine as long as you don’t rely on inline ASM and have ARM versions of any libraries you need.

You still might need to do some adjustments for specific behavior differences when it comes to concurrency and atomicity of operations. It didn’t surprise me to see that the previous attempt to make an ARM ReactOS didn’t support multi-threading, because trying to enable it was probably an unreliably buggy mess with race conditions all over the place.

Running another uarch is a whole new level of complexity vs just running on a different OS but with the same uarch, especially if concurrency is involved because translating from one instruction set to another can break atomicity assumptions that concurrency depends on to maintain coherency. You’d need to do thorough analysis of the code to determine where special care is needed, and even then, it won’t be trivial setting it up in a way that avoids deadlock because you have to understand what the threads are doing before you can say if it’s safe for one thread to wait for another (since they could end up waiting for each other).

Whereas running code meant for a different OS just requires implementing that OS’ API (and behaviour, possibly including undocumented behaviour some code relies on, which can vary from application to application, hence windows compatibility modes where they add a translation layer themselves). Not saying this is trivial, but compared to the above problem, it kinda is.

Not that ARM support is impossible, just if they manage that, it will be proclaimed loudly, not something that requires digging. If they don’t say it supports ARM, just assume it doesn’t.

Does ubuntu not support other desktops? I had little annoyances like that with fedora cinnamon, not quite enough to make me miss windows but enough that I’d notice them and wish I could adjust them, also I eventually learned that it was a desktop environment that heavily relied on some form of javascript, which likely explained why it sometimes couldn’t keep up with mouse updates. But then I tried KDE and it addressed all of my issues, plus some others that I didn’t even realize until I saw a better implementation, plus it’s able to maintain that realtime responsiveness cinnamon struggled with (and my machine is far low end).

It’s been a while since I used Ubuntu, and even then, it was just for school so I only really needed the terminal and didn’t care what the GUI was doing as long as it didn’t interfere with that.

Hope they update those to your liking soon in any case.

Is there a predictable difference between an exponential growth curve and a sigmoid curve before the linear growth section? Like I suppose you’d be able to measure the dropoff in acceleration as velocity reaches its peak, but given that this is also a random sample, sample noise would make that impossible to determine in real time.

I mean, it’s a % of people who use x chart, so the only way it won’t be sigmoid eventually is if it drops off as something else replaces it, but I don’t think looking at the chart will help predict where the chart is going any more than how well that works with stock prices.