Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
mirker
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
13 ms
·
91.
▲
by
mirker
4y ago
Well I’m not aware of them happening “much less” just that there are less categories of bugs. We will of course rarely have software written with and without GC so it’s not an answer we will find. Memory leaks happen very frequently in mana
92.
▲
by
mirker
4y ago
But when you think about it, what’s the benefit of driving? For reasonably short drives you don’t save much time with parking in cities and you don’t get any exercise.
93.
▲
by
mirker
4y ago
Memory leaks still happen if you have a misunderstanding of memory management, regardless of language. Keeping references to dead objects is one straightforward way to do it. C++ also has smart pointers so it’s usually a case of someone wri
94.
▲
by
mirker
4y ago
StarCraft mineral mining is interesting. 50 minerals per worker + 100 minerals per supply depot at some amortized rate. You have a dampening term on workers/mineral patch which saturates between 2-3 workers. It’s pretty linear till the
95.
▲
by
mirker
4y ago
Regarding 1) and following 2) you only need to make realistic-enough video that it passes as a person and is similar enough to the target person. For example, you can have a pretrained model (e.g., using zero targeted data) and search for s
96.
▲
by
mirker
4y ago
I’d think if your rate of spending is >$50/hour then that’s nearly-always a bug. The only reason this conversation is taking place is because serverless “infinitely scales”. Autoscaling physical instances has a max limit for similar
97.
▲
by
mirker
4y ago
Definitely. Housing in LA is not affordable. It would make more sense to build houses in the lowest cost of living area in California.
98.
▲
by
mirker
4y ago
Should I argue over magic, or…? Also, I don’t believe that’s the proper usage of that term. It would go like this “It cannot be magical, because math is logical.” Edit for maxdepth reply: The example is looking at any code for a model relea
99.
▲
by
mirker
4y ago
Linear algebra is fancy?
100.
▲
by
mirker
4y ago
I think the point is Walmart is able to pay at a low enough rate that it must be effectively subsidized by the government. Walmart then plays into it as a further benefit “they” provide.
101.
▲
by
mirker
4y ago
And also what industry would pay someone to solve that problem. Super hard problem of little industrial value.
102.
▲
by
mirker
4y ago
Well ya know, industry solved Fermat’s Last theorem numerous times but only managed to scribble it in the corner of their performance review.
103.
▲
by
mirker
4y ago
Why link to this one vs. huggingface?
104.
▲
by
mirker
4y ago
Seems similar to the Huggingface tutorial: https://huggingface.co/docs/transformers/parallelism . What’s different?
105.
▲
by
mirker
4y ago
It doesn’t have to be the same to achieve low loss. The date is not relevant, for example, for the image content. It can memorize a few images, but you can’t “compress” all of the internet’s images into text (if you can, make a startup w
106.
▲
by
mirker
4y ago
AWS had a paper in SOSP on verifying parts of S3 https://assets.amazon.science/07/6c/81bfc2c243249a8b8b65cc21...
107.
▲
by
mirker
4y ago
People knew how AI worked when AI meant “graph search.” But now solving chess is not AI and linear (and nonlinear) regression is…
108.
▲
by
mirker
4y ago
TPUv3 was supposed to use liquid cooling. Not sure if it actually is.
109.
▲
by
mirker
4y ago
It’s fine when the library is a tight abstraction. Unfortunately, ML libraries are leaky. Example: take a classification model and change the output dimensions without understanding the model.
110.
▲
by
mirker
4y ago
What are some of the things on the roadmap for the platform? Any immediate plans to close the command-line gap for TPU utilization, etc.? My overall impression is TPUs are pretty awesome but the software stack is still a bit hard to use com
111.
▲
by
mirker
4y ago
I make espresso, too. The problem is the espresso machine cost and quality impact the final product significantly. A high-ish grade machine is $700 or more. You can froth milk with less, but the coffee extraction usually is subpar. And actu
112.
▲
by
mirker
4y ago
As another commenter said, if you don’t care for quality, you can get pre-ground. Then, you just boil water and brew, which is a few cups of coffee, so potentially less amortized work. Cleanup is fair but it’s on the order of putting your h
113.
▲
by
mirker
4y ago
For coffee: French press + grinder is straightforward and has little maintenance. Machines have to be descaled.
114.
▲
by
mirker
4y ago
I’d say this goes with letters of recommendation, too.
115.
▲
by
mirker
4y ago
Yeah but that requires knowing control theory rather than “learning” it. Control theory is better if you know what you’re doing. ML is technical debt for sure.
116.
▲
by
mirker
4y ago
You should cross-reference the author’s alias with the NN512 web-page.
117.
▲
by
mirker
4y ago
Tin is normal (and also the reason the loaf is rectangular with a mushroom top rather than round): https://www.kingarthurbaking.com/recipes/classic-sandwich-br... For batter vs. dough: yes, they are not the same, but a
118.
▲
by
mirker
4y ago
If you make artisan bread with sourdough, it can easily last 3-4 days. Mine don’t grow mold until 7-9 days.
119.
▲
by
mirker
4y ago
There is “artisan” bread, made by hand, which is the round kind. Then there is “sandwich bread”, which is the rectangular kind. Ingredients: Artisan bread is usually just flour, water, salt, yeast (sourdough is just flour and water, too). S
120.
▲
by
mirker
5y ago
The API of Tensorflow and PyTorch is quite large. Even Tensorflow has a dialect specifically for TPU, so the notion of accelerator independence (especially when considering performance) has not been realized yet (though this is still a work
More ›