mariebks

5 ms·

1.
by mariebks 12d ago
Incredibly, Muzero didn’t even know the rules, it figured them out from starting with random moves: https://deepmind.google/research/alphazero-and-muzero/
2.
by mariebks 2mo ago
I could be wrong but I think the mesh hopping only works in lost mode, did you use that?
3.
by mariebks 3mo ago
GLM 5.2 is already between 4.6 Opus 4.7 Opus level based on Artificial Analysis aggregation. 4.6 Opus is about 4 months old at this point, so seems like open source is maybe 4-6 months behind. It could still take a year but seems closer to
4.
by mariebks 2y ago
I’m not an expert on this topic, but I think it comes down to cost and scalability. You have to construct a new project with custom specs for the exact site you’re on, and the permitting for a large environmental change is another drag. For
5.
by mariebks 2y ago
Hello fellow CI user!
6.
by mariebks 3y ago
Sodium Ion is looking promising for a cheaper alternative, with currently lower energy density than LFP (lithium iron phosphate) and NMC (high nickel li ion). BYD is trying to scale sodium ion, but based on analysis (I’m open to other point
7.
by mariebks 4y ago
Google audio search not only does speech to text, but can give you the song if you hum. I was with some friends who tried it outside in SF and it was a noisy environment on the sidewalk, and it was able to identify the song from their hums.
8.
by mariebks 4y ago
I think this is the big open question of evolution as a whole, and language and art can be seen as the evolution of ideas/memes as Dawkin wrote about in the Selfish Gene. Were these developments sufficiently iterative, or was there som
9.
by mariebks 4y ago
This is why I think concerts might be a killer app in the Metaverse. Imagine you could attend a super immersive and compelling concert experience with your headset and a special audio setup for $5, and for that price, millions upon millions
10.
by mariebks 4y ago
Can you link to some of your favorite talks/conferences/podcasts? Thanks!
11.
by mariebks 4y ago
https://www.teslarati.com/tesla-model-x-lidar-equipment-phot...
12.
by mariebks 4y ago
That’s exactly what they did. They used LiDAR to train their vision neural nets to accurately compute the distance of every pixel. https://cleantechnica.com/2020/08/03/tesla-achieved-the-accu...
13.
by mariebks 5y ago
Check out the app Sorted on iOS (not sure if on Android). You can set all your tasks for the day and all their lengths (great for timeblocking), a default buffer between each task, and then autoschedule all the tasks throughout the day. If
14.
by mariebks 5y ago
You could say that it’s optimized brute forcing. It seems to me that machine learning applied to combinatorial search problems of this nature cut down the search space massively by recognizing patterns of combinations that have a high proba
15.
by mariebks 5y ago
Sunlight through a window is much less bright according to lumen meters than when you’re outside. We have neurons that wake you up more and directly respond to bright light early in the morning. (paraphrased, source from Huberman Lab podcas
16.
by mariebks 5y ago
Tesla Megapacks.