LessWrong 2.0 Reader

The word for a drug that causes loss of memory is “amnestic”, not “amnesic”.  The word “amnesic” is a variant spelling of “amnesiac”, which is the person who takes the drug.

Thanks! Fixed now.

Oh yeah, I guess that could be a learning effect. When reading it I assumed the lack of need for repeating the numbers was just because the drug was wearing off.

This kind of study is really great because so much of what we know about short-term vs. long-term memory is based on just a few (admittedly, very convincing) lesion studies. I was always a little skeptical about the supposed clear distinction between short-term memory, long-term memory, and cognitive function, but this story really made me update in favor of this model! The fact that you could still perform pretty well on the spelling bee even without seeing the letters in front of you is impressive! I hope we see more LessWrong self-experiments along these lines (hopefully safely and under clinical supervision). :)  

I get conservation of expected evidence. But the distribution of belief changes is completely unconstrained.

Going from the class martingale to the subclass Brownian motion is arbitrary, and the choice of 1% update steps is another unjustified arbitrary choice.

I think asking about the likely possible evidence paths would improve our predictions.

You spelled it conversation of expected evidence. I was hoping there was another term by that name :)

In computer security, there is an ongoing debate about vulnerability disclosure, which at present seems to have settled on 'if you aren't running a bug bounty program for your software you're irresponsible, project zero gets it right, metasploit is a net good, and it's ok to make exploits for hackers ideologically aligned with you'.  

The framing of the question for decades was essentially "do you tell the person or company 

 with the vulnerable software, who may ignore you or sue you because they don't want to spend money?  Do you tell the public, where someone might adapt your report into an attack?  

Of course, there is the (generally believed to be) unethical option chosen by many "sell it to someone who will use it, and will protect your identity as the author from people who might retaliate"

There was an alternative called 'antisec': https://en.m.wikipedia.org/wiki/Antisec_Movement which basically argued 'dont tell people about exploits, they're expensive to make, very few people develop the talents to smash the stack for fun and profit, and once they're out, they're easy to use to cause mayhem'.  

They did not go anywhere, and the antisec viewpoint is not present in any mainstream discussion about vulnerability ethics.

Alternatively, nations have broadly worked together to not publicly disclose technical data that would make building nuclear bombs simple.  It is an exercise for the reader to determine whether it has worked.  

So, the ideas here have been tried in different fields, with mixed results.  

The linked post suggests that your assumptions about memory are wrong:

Interestingly, I asked her for two 2-digit numbers again toward the end of that hour, having no memory that I had already done this. She told me that she had already given me two numbers, and asked whether I wanted the same numbers again. I said yes (so I could compare my performance). The second time, I was able to do the multiplication pretty quickly without needing to ask for the numbers to be repeated.

He had training effects from multiplying the two numbers despite not having a memory of the first time he multiplied them. 

We will witness a resurgent alt-right movement soon, this time absent the institutional backlash that kept it from growing during the mid-2010s. I could see Nick Fuentes becoming a Congressman or at least a major participant in Republican party politics within the next 10 years if AI/Gene Editing doesn't change much.

Thanks! I'm personally skeptical of ablating a separate direction per block, it feels less surgical than a single direction everywhere, and we show that a single direction works fine for LLAMA3 8B and 70B

The transformer lens library does not have a save feature :(

Note that you can just do torch.save(FILE_PATH, model.state_dict()) as with any PyTorch model.

I should have added - Determine whether this is a modeling problem or a manufacturing problem.  If the model was sound but the physical screw was faulty, you'll need an entirely different response.

Shouldn't that be counting the number squared rather than the number?