I don't understand why a model has to follow the instructions. Don't get surprised when it shows its true color. Plus, do users (not the researchers) really check whether the response follows the instructions?

Models don't have to follow instructions, but during RLHF that is one of the things they are scored on so a premise of the idea is part of the model, but it's also balanced on accomplishing the end goal. They model may determine, correctly or incorrectly, that your rules suck and do what it thinks is best.

I'm a fan of user script, glad to see people are also developing new ones.

The uploads to greasy fork seem reduced these days.

When a string contains strange characters, how do you enforce the single consistent delimiter?

You probably haven't written diverse enough code.

If AI is so good at coding, why can't it be good at politics? We will elect AI in our next presidential election.

While you can use the hotline in private, you can't object to any matter in public.

From what I can tell, there are many issues that aren’t off limits to criticize on Chinese social media. In fact, recurring social media complaints are what spurred development of the hotline system.

It’s mainly complaints that are considered sensitive or destabilizing that are suppressed. This should sound familiar to those of us in the West. Germany actually goes farther by directly funding left-wing protest groups, as these are not considered destabilizing.

Exactly!

I installed a CO2 monitor at home, which reminds me of ventilation in winter when I keep windows closed and heating on.

Creating accounts should be allowed, but using an account could require age check.

People should be able to create an account at birth. Then when they grow up, they are ready to use the account. This way proves that the account owner is at least as old as the account.

Don't fight with AI. People who reported increased throughput don't verify AI output. Programs which are simply AI wrappers don't verify the output. If a serious programmer starts to design an algorithm that verifies AI output, the progress starts to slow down because you are doing something not needed if you don't use AI.

People here provide counterexamples to show this article is bullshit. Don't forget, fraud detection is about statistics. Outliers always exist.

Machine learning systems also learn your pattern. The article gives simple SQL rules. Don't dismiss this article as worthless.

Initially Cider was branded as a light client that opened much faster than traditional IDEs.

Now, ironically with so many extensions and LLM computing, users seem to forget that they chose Cider because of its lightweight.

Everything turns into the thing it was set out to replace.