thomasahle

35 ms·

511.
by thomasahle 4y ago
The argument above was that you can still hurt a bus driver psychologically by having them kill you. So it doesn't really matter that an ebike is heavier than walking.
512.
by thomasahle 4y ago
The argument above was that you can still hurt a bus driver psychologically by having them kill you. So it doesn't really matter that an ebike is heavier.
513.
by thomasahle 4y ago
You can also decompile a proprietary binary, ask GPT to rewrite it, and then re-release it as your own work. Somehow I feel the lawyers won't agree.
514.
by thomasahle 4y ago
Interesting. I have never seen one of those "diluted toblerones": https://www.ft.com/content/9ca7438a-a75e-11e6-8898-79a99e2a4... but they look ugly.
515.
by thomasahle 4y ago
> the theories have no utilitarian value If you've ever used a compiler, you've benefitted from "linguist theories" like regular or context free grammars.
516.
by thomasahle 4y ago
> Why is naming your product after a place that evokes characteristics you’d like to emphasize a lie? Try to replace "place" by "brand" in the sentence above. In Europe _places are brands_. Sure, in the US I might wan
517.
by thomasahle 4y ago
I'm replying to a comment about a hypothetical AI that can do math better than the mathematicians.
518.
by thomasahle 4y ago
Why does society have mathematicians right now? It's some mix of "the people find mathematics intellectually worthwhile" (public universities) and "we sometimes need to ask them to solve hard problems". If AI makes
519.
by thomasahle 4y ago
When they said the program "fixes" the image, they mean it undoes the photoshop. Hence the larger jaw.
520.
by thomasahle 4y ago
Discussion from last year: https://news.ycombinator.com/item?id=31064787
521.
by thomasahle 4y ago
How long before those beans have set you back thousands of dollars as well?
522.
by thomasahle 4y ago
> Markets are "efficient" at transferring money from "stupid" to "smart" people. Here "stupid" or "smart" is directly related to lower or higher in that information/power hierarchy.
523.
by thomasahle 4y ago
I agree a "Probability 101" article shouldn't use sigma algebras. It definitely cuts off 99.9% of the people that might benefit from the article. Not that sigma algebras don't have a place. If you are actually doing adva
524.
by thomasahle 4y ago
There might not be a simple topline objective everything is optimized for. More likely it's a linear combination of all kinds of dynamic business needs.
525.
by thomasahle 4y ago
The example in the blog was the original "Avatar date" conversation. The one I link is from someone else who tried to replicate it, and got an even worse gaslighting.
526.
by thomasahle 4y ago
And the suggested follow up questions: "How can I check my time machine?"
527.
by thomasahle 4y ago
My favourite conversation was this attempt to reproduce the "Avatar bug": https://www.reddit.com/r/bing/comments/110tb9n/tried_the_ava... Instead of trying to convince the user that the year is
528.
by thomasahle 4y ago
Unicode has these four: https://en.wikipedia.org/wiki/List_of_logic_symbols#:~:text=...
529.
by thomasahle 4y ago
Was I the only one looking for ⊥?
530.
by thomasahle 4y ago
I don't mind newspapers using paywalls. I even pay for a couple of newspapers myself. I just wish Google search would block all such newspapers. I'm probably not going to subscribe to any more newspapers, so all the walled results
531.
by thomasahle 4y ago
I generate my website https://thomasahle.com with a static Jinja template. Whenever I update it, it also generates a new resume and uploads that. Very convenient.
532.
by thomasahle 4y ago
> Fundamentally, I have never understood why a conventional neural network cannot learn self-attention. Here's how I think about it: Yes, you can learn everything with an MLP, since universal approximation and so on, but it is not e
533.
by thomasahle 4y ago
Nevermind, I guess it's "ImageNet-1K trained models" on which ViT gets 79.9% and the 90% is only when pretraining with ImageNet-22K. There are other non-attention based networks that get 90% too though: https://arx
534.
by thomasahle 4y ago
ViT gives you 90% top-1 accuracy on ImageNet: https://paperswithcode.com/sota/image-classification-on-imag... I don't know where they get the 77.9% number from. 75% is pretty bad. Similar to the 2015 VGG net as th
535.
by thomasahle 4y ago
For sure. I just wanted to point out the approach that arguably required "the least math knowledge". Also, OP was solving a degree four problem and wanted a closed form. For better stability just use `x = np.linalg.solve(M, y)`, o
536.
by thomasahle 4y ago
It's pretty hard for a nonexpert to come up with a good SQL injection. Doesn't stop it from being a security hazard.
537.
by thomasahle 4y ago
Edit: Should of course be without plusses in the matrix form: [x1^3, x1^2, x1, 1] [a] = [y1] [x2^3, x2^2, x2, 1] [b] = [y2] [x3^3, x3^2, x3, 1] [c] = [y3] [x4^3, x4^2, x4, 1] [d] = [y4]
538.
by thomasahle 4y ago
Sure, if you have more equations the degree, but that doesn't seem to be the case in the article.
539.
by thomasahle 4y ago
It's just ax1^3 + bx1^2 + cx1 + d = y1 ax2^3 + bx2^2 + cx2 + d = y2 ax3^3 + bx3^2 + cx3 + d = y3 ax4^3 + bx4^2 + cx4 + d = y4 As a matrix: [x1^3 + x1^2 + x1 + 1] [a] = [y1] [x2^3 + x2^2 + x2 + 1] [b] = [y
540.
by thomasahle 4y ago
Wow! I had no idea you could get fun projects like this into the NYT. How did you do it? :-D

More ›