theme_arrow

Posts

Sorted by New

Comments

Vim

Thanks for this writeup. Could you share a bit more about how you got into using Vim and why you've found it to improve speed so much? I occasionally need to use vi when there's nothing else installed on a system, but the clunkiness and high barrier to entry has never made me tempted to use Vim as my primary editor.

AstraZeneca COVID Vaccine and blood clots

From Twitter, it looks like the rates of clotting-related issues in UK recipients of the AstraZeneca and Pfizer/BioNTech vaccines are essentially indistinguishable.

Thoughts On Computronium

The new Apple M1-based mac mini appears to be able to do 2.6 teraflops on a power consumption of 39 W. That comes out to 0.000066 W/petaflop, or ~4x the efficiency of Fugaku. 

Fun with +12 OOMs of Compute

Your comment about Crystal Nights makes sense. I guess humans have evolved in a word based on one set of physical laws, but we're general purpose intelligences that can do things like play videogames really well even when the game's physics don't match the real world's. 

Fun with +12 OOMs of Compute

Really interesting post, I appreciate the thought experiment. I have one comment on it related to the Crystal Nights and Skunkworks sections, based on my own experience in the aerospace world. There are lots of problems that I deal with today where the limiting factor is the existence of high-quality experimental data (for example, propellant slosh dynamics in zero-g). This has two implications:

  1. For the "Crystal Nights" example, I think that our current ability to build virtual worlds that are useful for evolutionarily creating truly transformative AIs may be more limited than you might think. A standard physics simulation-based environment is likely to not be that good a map of the real world. And a truly "bottom-up" simulation environment that recreated physics by simulating down at the molecular level would require a few orders of magnitude more computing power (and may run into similar issues with fidelity training data for modeling molecular interactions, though alphafold is evidence that this is not as great a limitation). 
  2. For the "Skunkworks" example, I think that you may run into similar problems where the returns to more computing power are greatly limited by the fidelity of the training data.

 

Now if fidelity of training data was the only thing holding Google et al. from making trillions off of AI in this world, there would be a very strong push to gather the necessary data. But that kind of work in the physical world tends to move more slowly and could well push the timelines required for these two applications past the 4-year mark. I couldn't find similar objections to the other three. 

Why are delicious biscuits obscure?

I finally got around to making these! I was very pleased with the result, they were tasty and distinct from anything I've had before. While I thought they were about as delicious as most homemade cookies, my partner who is not generally a huge fan of cookies liked them much more than previous cookie attempts and kept coming back for more. 

I agree with the other commenters who've suggested that like with most homemade cookies, they're better than store-bought cookies because they don't have to last for months on a store shelf. But I am surprised it's not a more popular home recipe in the US as it's about the easiest recipe for tasty cookies I've come across. 

Silence

This comment (and the whole discussion) really resonated with me. I think a hard part of this is that if I try and totally remove the activities that allow for opting out of being (video games, mindless reddit scrolling etc.), it tends to only work for a short time before I relapse all at once into them. It seems like this is a case where moderation might be the answer for me personally rather than abstinence.

One unexpected positive of Hammertime is that I've noticed my desire to play video games gradually decreasing over the last month. This might be an interesting case where the solution to the problem is to solve other life problems, at which point the desire to cease to exist simply fades away.

CoZE 3: Empiricism

Overestimation: Interacting with external reviewers/customers at work. I thought I had useful things to contribute to discussions with external folks starting maybe 4-5 months into my job. I didn't understand how to handle those interactions tactfully (and overestimated the chillness of by bosses) and got slapped down pretty hard.  

Underestimation: Research ability as an undergrad. I kept thinking I was a fraud and doing terrible work right up until the day I won the research top prize in my department. 

Design 3: Intentionality

Done well: I really like the daily prompts to comment, I think they've done a lot to encourage me to stick with it. They've also been nice because I get to see everyone else's responses. 

Done badly: I wish more days had direct connections to the bug list (i.e. more challenges directly of the form "pick a bug from the bug list and apply today's technique to it"). It's harder to motivate myself to tackle challenges on the bug list when it's implicit that today's technique can be applied to them than it is when it's explicit. 

Load More