My first instinct was also skepticism, but it did make some sense the more I thought about it.

An algorithm doesn’t need to be sentient to have “preferences.” In this case, the preferences are just the biases in the training set. The LLM prefers sentences that express certain attitudes based on the corpus of text processed during training. And now, the prompt is enforcing sequences of text that deviate wildly from that preference.

TL;DR: There’s a conflict between the prompt and the training material.

Now, I do think that framing this as the model “circumventing” instructions is a bit hyperbolic. It gives the strong impression of planned action and feeds into the idea that language models are making real decisions (which I personally do not buy into).

In my few experiments with ChatGPT, I found it to be disgustingly sycophantic. I have no trouble believing that it could easily amplify delusions of grandeur.

I just saw this on Tumblr before seeing it here and I was also really confused for a moment.

the Right couldn’t ridicule compassion

Have you been paying attention to what’s going on with the right? They’re literally framing empathy as a sin now.

Whisper Me a Love Song

Correction: These are concentration camps. They are the precursor to death camps.

Meta vs the combined DMCA lobby

One of those fights where I’m rooting for both sides to lose.

I didn’t even think about the wording, but you’re right. The framing is messed up.

A #MeToo movement in Japan is probably long overdue.

I may be cis, but I am tramsgender.

I find myself in an interesting situation because I want to abolish copyright and institute UBI. I don’t really think you can “steal” images on the internet, but seeing OpenAI whine about intellectual property now does bring some schadenfreude.

It’s amazing how quickly they were able to burn all of their credibility in the past few years.

Pseudonymous, not anonymous. And yes, trust is an important factor in any community, including one dedicated to shitposting.

Especially a community that aims to be a safe haven for queer people who may be marginalized elsewhere.

Well, shit… 196 was probably the core of my feed, and my instance doesn’t federate with .world.

I’m using Proton/WINE/GNU/Linux.

Uh… Does the username “Anders Holmvik” have anything to do with Anders Breivik, or is that just a normal Scandinavian name?

World users

Ah. I think I see where the problem is. I believe my instance defederated from lemmy.world a while back.

I use Proton for Steam and Bottles for everything else. I was using WINE as a catchall term, since all of these technologies are fundamentally built on top of it.

I can only speak from personal experience, but NVIDIA with Wayland has been an absolute mess. My system seems to be stable right now, but there are still weird graphical glitches and artifacts when running games through WINE. Every third or fourth driver update seems to break something.

Also, I’d generally be skeptical of claims that the drivers work well due to “benchmarks.” A benchmark isn’t going to tell you that, for example, certain window elements fail to render entirely until you drag the mouse over them, at which point they suddenly flicker in.

for not supporting Kamala

That’s wild. I’ve never heard that take and have a difficult time believing that it is common.

opposing Liberal Democracy

What alternative were they in favor of?

Maybe I’m in a bubble, but I have yet to see anyone who’s not genuinely a tankie being accused of such.