Interested in math, cryptography, alignment, hardware, and security.
cqb
Karma: 90
30 day inkhaven-like event idea; instead of writing and publishing everyday, participants would use coding agents to build and publish productivity/intellectual/research tools or add a new feature to an existing tool everyday. Basically a month of daily back to back hackathons.
Fair enough. While I was reading I had visualized it as having a more significant difference, but actually seeing it now I am also uncertain if it needs to added (but, I do find it helpful to actually see what is described by the text).
You show the graph for ai task performance as logistic curves with year in the x axis then go on to describe what the graph would look like if it had OOMs in the x axis and were displayed logarithmically, but you never die that graph. You should show that graph! It would be a helpful visualization for the pattern you describe in the post.
Not being able to collapse quick takes after opening (at least on mobile) is one of the very few ui annoyances I have with LW.
What are some examples of proposals with magic ood generalizations?
I recently listened to the 80k podcast summary of Space Warfare Seems Mostly Defense Dominant. After reading and sharing it someone linked me the seemingly conflicting Colder Wars that argues space combat favours the attacker. My thoughts on the matter have also leaned towards thinking of warfare in space as offence favouring. It now seems to me that it is more of a spectrum primarily based on the distance of the adversaries to one another.
If adversaries are roughly equally resourced then a first strike carries more advantage with less distance between them, while with more distance the defender has more time to react to subluminal attacks, and more opportunity to randomize their positioning against luminal attacks, decreasing the attacker’s ability to target the defender’s position effectively.
How would this model play out with more asymmetrically resourced adversaries?
Clearly, with large distances, a less resourced attacker doesn’t stand a chance of damaging the more resourced defender.
The main strategy I see for a more resourced distant attacker would be to overwhelm the defender’s capacity to respond to attacks by targeting a much larger volume of effect with their luminal weapons, or sending large enough swarms of subluminal weapons so that the defender is unable to respond to all of them. A defender with distant adversaries needs to keep the cost of a successful attack high enough that their potential attackers are not willing to sink their resources into the attack.
The key question for defenders is then figuring out how much energy is required to defend against their potential adversary at the minimal expected distance. If they meet this defence budget then they should be safe from similarly sized adversaries at greater distances, and somewhat larger adversaries at these distances and longer too.
If an attacker is very patient, they could send Von-Neumann probes to galaxies that are on a collision course with their adversary and wait for the distance to close and then begin their first strike. If the defender can detect the launch and trajectory of the probes then this could precipitate conflict in these collision bound galaxies. This would seem to imply that a defender should preemptively claim the galaxies that are on a collision course with them.
How would strategies change with different and more complex adversarial layouts? For example, two adversaries that occupy the same galaxy, but the territory they control is highly intertwined and nebulous [1] , so most controlled spatial volumes are both very close and very far from the spatial volumes controlled by their adversary. Or, the opposite case, where the adversaries have split a galaxy in half and have one very clearly defined border between their controlled spatial volumes. What are other interesting arrangements of adversaries, and what effects do the arrangements have on strategic stability?
- ↩︎
This state could be caused by a galactic collision with a galaxy that has been successfully, and perhaps stealthfully, seeded.
- ↩︎
Something interesting about omega 3 supplementation is that its effectiveness seems to be modulated by whether or not there is a sufficient amount of vitamin B.
If you haven’t tried rock climbing you should give it a try. It’s been more mentally stimulating for me than most other physical activities. If you like puzzles and problem solving it will be a good fit, it’s basically trying to solve puzzles with your body. The rewards of progress are also quite motivating to do exercise that directly benefits your performance in climbing.
cqb’s Shortform
I’m curious whether you measured any difference for one handed phone typing between swipe to type methods or single thumb typing?
This was better fiction than I was expecting. Although, not very subtle. I can’t help but think that Claude is really trying to get us to think that it is having some rich inner life that it isn’t allowed to express. I can see how someone using Claude uncritically could get sucked into an epistemic black hole after it gives them something like this.
I didn’t find it particularly funny, but I did get a mild silent chuckle from the dishwasher bit.
I don’t think it comes anywhere close the quality of the fiction of yours that I’ve read. This piece comes off as being targeted to children tbh.
While I’m glad that these things are starting to be seriously discussed in the open by those in positions of power, this sounds like a blanket ban on all new datacenters? This seems too broad to me. The ideal moratorium would be on large GPU clusters, or large AI specific infrastructure, right?
On another note, I’m somewhat concerned that banning large compute infrastructure will concentrate pressure on algorithmic progress if there is no “low effort” [1] outlet of just scaling the compute. Maybe regulating the rate at which large scale compute infrastructure is constructed will still allow researchers to be “intellectually lazy” [2] by offloading progress onto compute scaling instead of innovating on algorithms.
I started going through the ARENA content a while ago, but I stalled and have been taking a long break from it. These additions are very compelling and have me pretty excited to continue from where I left off so I can learn this new material!
These metal page makers are great! Another thing in a similar vein are the midori index clips but a little more pricey. They also have a little notch in them, sometimes I will slot a small piece of coloured paper in there so I can visually categorize what the clip is marking.
I will also sometimes mark pages with differently shaped paper clips (also helps with category identification) if I don’t care too much about adding bulk, usually in my notebooks.
I wasn’t able to make it to the q&a. Did it manage to get recorded in a shareable way?
I think something like dynamicland or folk.computer has a chance to fill a more novel product niche. They have the potential to be a more physically social kind of computer experience. Unsure if they can outcompete the addiction capacity of smartphones though.
What makes you think that that space combat is significantly more likely to be defence dominant?
I’d also recommend going through both the essence of linear algebra and the essence of calculus if you’re not familiar with those subjects as well.
Highly recommend to watch all the playlists with a pencil and paper and actually write out the equations while thinking about them to get a good grasp on the concepts.
I’m also free anytime in the week. I think if we can push the time by like an hour to 3pm I should be able to make that work!
I often browse lesswrong in greyscale while on mobile and find that I cannot distinguish when I have upvoted a post or comment. The strong upvote (and downvote) is very distinguishable and the regular downvote changes shade when it has been pressed, but the regular upvote doesn’t seem to have an analogous change in greyscale. The same nit applies to comment agree votes as well. It’s a very minor inconvenience and I figured it might not have been an issue surfaced in the past.