faul_sname
Fuck around once, find out once. Do it again, now it's science.
No bio...
User ID: 884
faul_sname
Fuck around once, find out once. Do it again, now it's science.
17d ago
Question as phrased, blue, because lots of people will look at that question and press blue.
More evil variant:
You are told that you will be presented with two buttons, one red and one blue. If you press the red button, you will survive and no additional people will be told about the buttons due to your actions. If you press the blue button, the question will be presented to two other random people in the world (unless no people remain who have not been asked the question). Once this process completes, the votes will be tallied. If strictly more than 50.0% of the people asked pressed blue, everyone survives. Otherwise, everyone who pressed blue dies.
What's the ethical thing to do here?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
19d ago
If you have lower ability than your community then you will vote for more redistribution.
Support for socialism increases with education and SES, no?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
20d ago
Than what now?
faul_sname
Fuck around once, find out once. Do it again, now it's science.
20d ago
The Department of Justice has filed a Rule 62.1 motion on behalf of the National Park Service, et al.1 asking the district court for an indicative ruling that it would dissolve its preliminary injunction against the White House ballroom construction project if the D.C. Circuit remands for that purpose. In other words, the DoJ says that the assassination attempt on Saturday shows that the district judge's analysis was clearly mistaken, and asks that the district judge formally state that he would lift his own injunction if the D.C. Circuit sent the case back to him for that purpose.
The motion opens:
The National Trust for Historic Preservation” is a beautiful name, but even their name is FAKE because when they add the words “in the United States” to the National Trust for Historic Preservation, it makes it sound like a Governmental Agency, which it is not. In fact, the United States refused to continue funding it in 2005 because they strongly disagreed with their mission and objectives. They are very bad for our Country. They stop many projects that are worthy, and hurt many others. In this case, they are trying to stop one that is vital to our National Security, and the Safety of all Presidents of the United States, both current and future, their families, staff, and Cabinet members. They were asked by the United States Military not to bring this suit because of the Top Secret nature of the important facility being built. They were shown detailed plans and specifications of this knitted, unified, and cohesive structure by Top Officers and Leaders in both the Military and Secret Service. But this did not deter them because they suffer from Trump Derangement Syndrome, commonly referred to as TDS, as noted by Democrat Senator John Fetterman, of Pennsylvania, and are represented by the lawyer for Barack Hussein Obama, Gregory Craig.
Putting aside the bit where this looks like a Truth Social post rather than a legal filing, the legal strategy of the Trump Administration is interesting here - as far as I can tell, they are making the argument that because there is a legitimate security concern, they should be able to make these changes to the White House without having to go through the usual NCPC review, NEPA review, Congressional authorization, etc. This would
1 where the "et al." is the Executive Office of the President, the White House Chief of Staff, the Office of the Executive Residence, and the White House Chief Usher
As usual, Gwern has the canonical post on this topic.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
24d ago
A lot of people getting put on the street
It's getting harder to stay off the street
He comes home every night past tents on his street
And wonders how much longer till he's out on the street
faul_sname
Fuck around once, find out once. Do it again, now it's science.
26d ago
AHA. Turns out I was not wrong a year ago when I said people would freak out a bit once they realized that LLMs could do this, just early.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
26d ago
The easiest way to say this without being canceled is "if you have a heritable propensity to abuse others, your family members probably share that propensity"
faul_sname
Fuck around once, find out once. Do it again, now it's science.
26d ago
The reason I said "if you instruct Claude to use the programming env" is that Claude will generally do things similar to those that were evaluated well in the past, and most chess-like-evals would have forbidden tool use or anything else that human players wouldn't consider "fair play". I expect "always consider what tools you have available and make use of them where it makes sense unless explicitly told not to" in your user instructions will work so that you just never run into this in practice.
Bluntly, I don't think it matters how the board state is represented, as long as the answer isn't "Claude is trying to reconstruct the entire board state from the move sequence".
FWIW I tried the prompt
Play good chess.
d4
and Opus 4.7, at various points in the opening, dumped a snapshot of the board state into the chat.
Not playing the full game because Claude spent 20 minutes thinking and writing janky minmax code after blundering before hitting the compaction limit then erroring out, then on the second attempt spent another 15 minutes thinking and almost hit the compaction limit but you can see that it does in fact use tools.
Anyway, to answer the question:
Firstly, why should I as the user have to prompt the AI to make a program to ensure that it doesn’t go off the rails? Why can’t it figure that out for itself?
The AI has no memory. Every conversation is a fresh new world. As a rule of thumb, I expect AI to significantly outperform me at anything I've never done before, but that for any task that hasn't been the subject of absurd amounts of RL (and some tasks that have), I'll very quickly be able to identify the places that AI is likely to fail and steer it around those pitfalls. Because I can learn, and the AI can't.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
28d ago
An LLM with access to a sandboxed coding environment (and instructed to use it) will generally not make illegal moves in a chess game.
faul_sname
Fuck around once, find out once. Do it again, now it's science.
28d ago
The word "good" in "good continuous learning" is doing a lot of work. Sample efficiency improves with model size, so it's possible that current models are just too small for good continuous learning, rather than there being some theoretical piece that's missing.
Alright, variant for you:
Instead of being drawn from the global population, additional people are drawn from the set of people who have ever argued with someone about (explicitly) decision theory on the internet.
(also, I think I found a situation where FDT performs worse than naive CDT - naive CDT + not arguing about decision theory will always achieve outcomes that are strictly not worse than FDT in this scenario)
More options
Context Copy link