krasin

12 ms·

121.
by krasin 2y ago
> It's an interesting observation, but I couldn't replicate this by taking a photo of a Google Images search for "north face logo" Low-level artifacts for screen photos are different, so whatever neural net messes thi
122.
by krasin 2y ago
> Small correction: Scanning tunneling microscopes can do that. Thanks! You're correct.
123.
by krasin 2y ago
Scanning electronic microscopes can see and/or manipulate individual atoms. So, the answer to your question is simple: they took a look from the side and confirmed that it's 1 atom thick. 1. https://education.mrsec.wisc
124.
by krasin 2y ago
Here is the tutorial how to recreate an ALOHA cell: https://docs.google.com/document/d/1sgRZmpS7HMcZTPfGy3kAxDrq... The robot arm kits are available commercially: https://www.trossenrobotics.com/al
125.
by krasin 2y ago
It's already open-source; most of it anyway: 1. https://github.com/tonyzhaozh/aloha 2. https://aloha-2.github.io/ 3. https://github.com/tonyzhaozh/aloha/tree/main&#x
126.
by krasin 2y ago
> or tons of energy will be wasted. There are good potential industrial loads for that intermittent extra energy (nuclear or solar): desalination, CO2->methane, air->ammonia, etc.
127.
by krasin 2y ago
Another discussion from today: https://news.ycombinator.com/item?id=40045252 (77 comments).
128.
by krasin 2y ago
> Historically, this has been hydro from surrounding states (like Washington, but also from Canadian BC). I'm not sure what it is today given current water challenges. And coal from Utah. That's still huge at night.
129.
by krasin 2y ago
And desalination.
130.
by krasin 3y ago
Normally, I would also prefer reading, but for special effects, I would rather see them on video.
131.
by krasin 3y ago
I would be interested in some links!
132.
by krasin 3y ago
> The point I'm trying to make is that to do them well, to have the bugs worked out, to have them reliable enough for production use, etc. is much more difficult than one would think. Oh yeah. Even if it's not scam, it is goi
133.
by krasin 3y ago
> In addition, they talk about abandoning digital controllers because of cost and going for analog only, which makes my bullshit detector go off. This is only for driving galvo motors. And analog makes a lot of sense for that. Complexity
134.
by krasin 3y ago
> It's petty, but also completely within their right Not necessarily. The right to refuse business usually does not apply to monopolies. A good historical example is AT&T and its telephony service. It's entirely feasible th
135.
by krasin 3y ago
I seriously considered using Grayskull. My desire was driven by the interest in NeRF and Gaussian Splatting, for training whose the Grayskull arhitecture (at least, on paper) could considerably outperform NVIDIA. I decided against it for th
136.
by krasin 3y ago
Texas is indeed a weird place to start deploying small nuclear reactors: electricity is quite cheap, as there is abundant wind and solar. I would have thought that isolated Alaska settlements would be much better, especially, if SMRs also p
137.
by krasin 3y ago
Combustion engines mitigate that problem by having multiple cylinders. An eight-cylinder car engine runs much smoother than a single-cylinder scooter.
138.
by krasin 3y ago
This is a ~400 meters segment; the method is quite lossy due to wireless power transmission and also encumbered with a pending patent. I also suspect that road maintenance will be much more expensive, once there are smarts inside. A time-pr
139.
by krasin 3y ago
AGPLv3 is a fine license too. But most of the models nowadays come with bullshit licenses, like Llama 2 with its "acceptable use policy" enforced by the license: https://ai.meta.com/llama/use-policy/
140.
by krasin 3y ago
The license is a proper open-source one: Apache 2.0. Thanks, AI21 Labs.
141.
by krasin 3y ago
> Angular Well, Angular is kind of dead now (or rather - will be). See https://twitter.com/sarah_edo/status/1770478763253379488 > Today we have some exciting news! We're merging frameworks! Angular and
142.
by krasin 3y ago
These two links could be a good start: ALOHA-2: https://aloha-2.github.io/ RT-X: https://robotics-transformer-x.github.io/
143.
by krasin 3y ago
Finetuning LLMs is currently the most promising way for next-gen robotics. One of such works (PaLM-e) among other things measured the impact of finetuning on general purpose tasks: https://twitter.com/DannyDriess/status
144.
by krasin 3y ago
> I don't think the market crowding here is great for anybody. In the end, the users will win from this crowding, I suppose.
145.
by krasin 3y ago
Previous discussion (1 month ago, 292 comments): https://news.ycombinator.com/item?id=39263854
146.
by krasin 3y ago
Sugar content grows, though => obesity is much more widespread.
147.
by krasin 3y ago
Just like Canadarm ([1]) on ISS considerably simplified docking, I can see a similar approach working for lunar landing: slow down to below 5 m/s, get caught by a robotic arm and gently put into a good spot. 1. https://en.wi
148.
by krasin 3y ago
I would like to quote this comment from an earlier discussion on HN: https://news.ycombinator.com/item?id=39544442 . > Perfect crystals. Also, the proof-of-concept drug was ritonavir, and it's nearly impossible to c
149.
by krasin 3y ago
This sounds a lot like "brick layers" by CNC Kitchen: https://www.youtube.com/watch?v=5hGm6cubFVs - I believe that something like that will be a must for future 3d printing slicers, as it increases mechanical stre
150.
by krasin 3y ago
> But what that also means is that this is closer to generative AI than to objective measurements. If the image to depth estimation goes very wrong, it might hallucinate shapes that aren't there. But people do that all the time too.

More ›