Obviously the judge who sentenced him expected him to die in prison. Yet legally, he will be set free in the year 2980. Perhaps the law will have to be reinterpreted to argue for the judge's intent: that the government wants him to die behind bars. But if he duly serves the time he owes, we can’t arbitrarily change the law to prevent his release, right? Nevertheless, after stewing in prison for almost a millennium, do we think it’s a good idea for him to rejoin society?

This sounds like you don't understand how the law actually works. In the real world prisoners are frequently released without serving all the years that they are sentenced to serve. You don't need to reinterpret the law to release someone earlier, you just need to apply the law as the law is actually applied in practice.

Parole is a thing. 

It seems like there are things that might have happened:

1. ChatGPT, failed to use it's internal API correctly and while it attempted to delete memories it didn't succeed and just reported that it deleted memories because it tried to do so and is unware that it failed.
2. ChatGPT did not try to delete any memories but reported that it deleted the memories.
3. There's a misunderstanding between the two of you about what you mean with memories. 

While 1/3 are bad they are not schema and only 2 seems to be scheming.

Without having the chat log it's pretty hard to tell from the outside what this is about. 

I don't think "expending political influence" is a good model. If you manage to work with anyone to get any policy passed, in the process of doing that work you will build relationships that help you pass other policies in the future.

Currently, US healthcare policy is at a place where there's more potential for change then there was for a long time.

Basically, you are saying that you don't know what you are talking about. On the other hand, the person who started this post does know what they are talking about from talking with people in the AI governance space.

For effective political action, it's useful to take insider information about how the process works seriously. 

There are not millions of rats/EA. 

Generally, if you want to make a political impact by being politically active it makes sense to have a theory of change and pick actions based on that theory of change. If you just randomly copy the kind of things other people who want to make political change do like going to protest and making small donations, the impact is likely not going to be very large. 

This reminds me of a discussion I had with someone from East Europe going to a government protest. I asked her what she was protesting and she said that there's a complex situation and no good English language source that describes it. Writing an article (or talking someone else into write an article) for the Guardian's Comment is Free to explain the decision would have done much more to affect political change than having another body at the protest, yet she did the easy thing of going to the protest instead of doing the EA thing of getting the article written.

A key aspect of typing is that it allows other skills to be expressed. If I want a given software written, I need both my typing and my programming skills. Employing someone who has no typing skills to do the programming for me is a bad idea. On the other hand, if I employ some with programming skills equal or better to mine it works. For the quality of the end product, the programming skills are a lot more important than the typing skills.

Do you have an idea about what kind of skills your CAD skills allow you to express that a random person you hire with CAD skills might not possess?

That's not true, the US is not equally governed by Republicans and Democrats at the same time with Republicans and Democrats always acting the same. The reason that this post exists is likely downstream of how the Trump administration manages hiring for positions. 

While not specified in this post, and me not knowing the exact ground reality, it's possible that the Biden administration was more technocratic about their hiring decisions. It's certainly possible that a future democrat administration also cares a lot about this, but also possible that they don't.

Eliezers post about how politics is the mind-killer is about the idea that if you talk in a non-political post about politics, you make it a lot harder for your reader to engage with the non-political part of your post. 

The question about whether or not to donate money to political candidates is inherently a political question. 

DanielFilan is essentially saying that even so, he thinks Trump is bad, people should still be wary about making certain donations for Democrats to oppose Trump. 

Taking government positions to influence AI is being politically involved. The political impact it has is likely a lot higher than that of a small donation.

Donations to SuperPACs and donations to political candidates are not exactly the same thing.