• 0 Posts
  • 35 Comments
Joined 2 years ago
cake
Cake day: March 4th, 2024

help-circle
  • I dislike organizational psychologists. The whole field. Every time a problematic hr department does something problematic using ‘researched methods’, the researchers all cry foul and say ‘that’s not how you’re supposed to implement <policy/test>!’ But do you not see your own role in legitimizing this behavior? All the research into how to run a happier bee colony is going to be used to justify mistreatment and discrimination by bad actors. Every test and every policy can be maligned, and providing citations just makes that easier. The research field is itself a tool for maintaining capital.



  • My question is why these harnesses are even necessary. The cloudflare pipeline is not specific to any codebase, it is just secret sauce they added themselves that increases the costs dramatically. Cloudflare is not an AI company though, Anthropic is, and openai and anthropic have spent tens of millions on signing bonuses for all of the most competent AI researchers in the field.

    Why is it cloudflare’s job to make the model useful? Why doesn’t the model do what it says it will without multiplying the token burn rate 5-10x? Why not ship a harness developed by the ai experts, if a harness is truly necessary? The idea of adverserial machine learning is more than a decade old, it’s not like cloudflare stumbled on a new concept.

    I believe this is just another attempt to hide the true cost of inference.


  • That cloudflare blog post already spilled the beans: it’s just unreliable. The same task could be repeated three times and come up with three different answers, sometimes refusing outright, sometimes failing, sometimes returning false positives. Only a fraction of runs turned out anything useful, and that’s only after running a separate instance of mythos to sort through the trash, and then multiple more verification runs.

    So, the most expensive model, burning the most tokens, you still need two instances and multiple runs through the same underlying task, and it may give you an exploitable bug. My understanding from cloudflare’s blog post is that their complex harness is entirely in-house, so I really think most of anthropic’s partners are having an even worse experience sorting through mythos trash.

    My feeling is that there is a diminishing rate of return on token burn rate. I also believe increasing the complexity of models makes it harder to set boundaries and control output.

    Also, most of the bugs so far have come down to not using basic OS safeguards or the attacker already having access to your computer. They are important threat vectors that need to be addressed, but they are types of vulnerabilities we’ve known about for decades and built protections around.



  • I used vapes to quit cigarettes, and then patches and gum, and now it’s been close to 3 years since I last smoked. Changing the flavor depending on my mood was I think really good to dissassociate tobacco taste from the nicotine addiction. In my area there was a shop that made everything to-order, had a bunch of crazy recipes that changed every week and you could step the nicotine down to any %. He shook it up right there, it was a chill place too with games and vending machines while you wait. No judgement, very personal. Every time I came in he constantly tried to ask if I qualify for some discount or another, ‘are you a member of the local church?’, ‘are you a firefighter?’, and I’d have to keep saying no until he just picked a random one to apply lol.







  • I think people like to see marketing as an art when in reality it is more of a science. Ads are engineered to be cost effective, and the design simply follows from that incentive. By taking up any of your attention, even if it’s in the background, even if it’s a small patch on an athlete’s uniform, it will help the product sell. If it didn’t, advertisers wouldn’t have put it there.

    They have more data than you, they are smarter than you, and all they care about is selling their product. You are a tool to boost profit, even if you don’t directly give them money, your attention is all they need.

    Like, this comic is actually a good example of how an anti-consumption response to an ad can actually help sales. If she never knew about Dragon Marrow, she wouldn’t be opposed to it, but you will inevitably just give Dragon Marrow free marketing every time you talk negatively about them. You may drive a majority of people that hear you complain away from it, but maybe one or two of your friends decide to try it anyways, people who would not have heard of it until you complained.

    Attention is the goal, not a specific emotional response. You can’t outsmart an ad by watching it.



  • Yeah spot on. ‘Granite’ is a wide category of rock. The Vietnam memorial is also apparently granite, I just learned that today. It looks nothing like the reflecting pool, it’s much darker, almost black. The specific types of granite, maybe even the specific quarries, was chosen by each architect for it’s texture and color. Congress paid extra to keep the original granite floor while replacing everything else about it in 2009.

    It will never be restored, I don’t think it’s possible. Our generation will have been the last to be able to see the original intended art piece.





  • No bid contract to a friend of Trump’s in the hotel pool industry. The idea stemmed from Trump seeing the park service scrape algae off the top because of ongoing filtration issues. That’s it. He just wants to hide the algae. It’s so dumb, and as far as I understand it, pool paint is nearly impossible to remove.

    Hey did you know that darker colors reflect less sunlight? They turn a substantial portion of light into heat; I wonder if that will affect the algae. I also wonder how he ever thought something on the bottom of the pool was supposed to hide the surface. That’s not how occlusion works.