Late in 2019, I, like many of my rationalist friends purchased the parts for and assembled a genuine, bona fide LUMENATOR™️ - a device for greatly increasing the brightness of your home - according to the original specification. To me, lumenators are the quintessential application of the More Dakka mindset:...
Authors: Roy Rinberg, Adam Karvonen, Alex Hoover, Daniel Reuter, Keri Warr Arxiv paper link One Minute Summary Anthropic has adopted upload limits to prevent model weight exfiltration. The idea is simple: model weights are very large, text outputs are small, so if we cap the output bandwidth, we can make...
Late in 2019, I, like many of my rationalist friends purchased the parts for and assembled a genuine, bona fide LUMENATOR™️ - a device for greatly increasing the brightness of your home - according to the original specification. To me, lumenators are the quintessential application of the More Dakka mindset: when you face a problem that responds positively to a little bit of X and responds even more positively to larger amounts of X, you don't just stop applying X once you feel you've done a reasonable amount of it, you add more and more and more until your problem goes away or X stops working. I built a lumenator not for... (read 742 more words →)
we generally recommend using this Google Docs add-on to convert your Google Docs drafts (formatted as you like) into Markdown syntax.
btw google docs supports this natively now! Select some text, press Option+/ and search for "Copy as Markdown".
Authors: Roy Rinberg, Adam Karvonen, Alex Hoover, Daniel Reuter, Keri Warr
One Minute Summary
Anthropic has adopted upload limits to prevent model weight exfiltration. The idea is simple: model weights are very large, text outputs are small, so if we cap the output bandwidth, we can make model weight transfer take a long time. The problem is that inference servers now generate an enormous amount of tokens (on the order of ~1TB tokens per day), and the output text channel is the one channel you can't easily restrict.
Nonetheless, in this work we find that it’s possible to dramatically limit the amount of information an adversary can send using those output tokens. This... (read 2304 more words →)
Has anyone tried using grow lights such as this one for bedroom wake-up lighting?
Specs:
Claude tells me that 1786 µmol/s PPF and 2.8 µmol/J PPE = ~100K lumens, ~50K lux @ ~6'. This sounds really compelling to me, maybe too much? Doesn't specify CRI. I'm thinking I'll need to mount a diffuser below it so that I'm not staring at point sources when I wake up.
Join the manifold.markets group here: https://manifold.markets/group/vancouver-games-night?referrer=zoli
For our inaugural meetup, we're going to sit down, play some games, answer some trivia, and even do some social deduction and scheming, but with one big twist: every outcome will be represented by a play-money prediction market on https://manifold.markets/!
Everyone will start the night with 500 Mana, the currency of manifold markets. You can gain mana by correctly predicting outcomes, or by day-trading, or any number of other strategies. Whoever finishes the night with the most Mana wins a cool and rare prize!
Please note: you do not need to know anything about prediction markets to participate! We'll dedicate the start of the event to getting familiar with https://manifold.markets
This event is inspired by... (read more)
Update: I built it! https://www.lesswrong.com/posts/p869mKDGL3onYsGYt/lumenator-2-0 It's pretty good :)