hexomancer

8 ms·

31.
by hexomancer 1y ago
Something never thought I would say: google AI previews. They actually helped me a lot during Iran's internet shutdown last week. I wrote a blog post about it: https://ahrm.github.io/jekyll/update/2025/06
32.
Google AI previews helped me in Iran's internet shutdown of 2025(ahrm.github.io)
3 points by hexomancer 1y ago|0 comments
33.
by hexomancer 1y ago
What’s the alternative? Live under the delusional islamic theocratic dictatorship forever? If you think the islamic republic can be replaced through peaceful protests you have another thing coming, they have already killed thousands of prot
34.
by hexomancer 1y ago
I have no idea but I sure hope so.
35.
by hexomancer 1y ago
Iranian here. Most Iranians use Telegram or Whatsapp, both of which are blocked right now and can only be used using VPNs (they have been blocked for years, though whatsapp was unblocked a few months ago, but it is blocked again after the I
36.
Show HN: Visualize nested list properties in the CodeLLDB debugger(github.com)
2 points by hexomancer 1y ago|0 comments
37.
by hexomancer 1y ago
Yes, every token is expanded into a vector that can be many thousand of dimensions. The vectors are stored for every token and every layer. You absolutely can not fill even a single research paper in 2 GB much less an entire book.
38.
by hexomancer 1y ago
Here are some objective issues with this page that I don't think is really up to taste (honestly these are so obvious that I assume you viewed the website on mobile which is fair, I never used the mobile version. Because I don't t
39.
by hexomancer 1y ago
It's funny that the people who designed this monstrosity of a web page feel qualified enough to advice other people about design.
40.
by hexomancer 2y ago
This is false. The const of producing a single token is linear but the cost of producing an entire sequence of length N is O(N^2) still (which is always what we meant when we talked about quadratic cost not the cost of a single token).
41.
Bigscreen Beyond 2 [video](youtube.com)
3 points by hexomancer 2y ago|0 comments
42.
by hexomancer 2y ago
When I read the title I thought maybe they implemented a typescript to binary (instead of javascript) code compiler that speeds up the program by 10x, it would also have the added benefit of speeding up the compiler by 10x! I don't thi
43.
by hexomancer 2y ago
Do you not see how your argument could be applied to steam engines putting human laborers out of work? Or computers putting (human) calculators out of work? Do you think inventing the steam machine or computers was a mistake too?
44.
30 Weird Chess Algorithms: Elo World [video](youtube.com)
4 points by hexomancer 2y ago|0 comments
45.
by hexomancer 2y ago
Isn't the fact the P40 has horrible fp16 performance a deal breaker for local setups?
46.
by hexomancer 2y ago
Here is how I make sense of it (I have no expertise in this subject, please feel free to correct me if I am wrong): I think when the model is pretrained on the internet, it does gain most of the skills required to do mathematical reasoning,
47.
by hexomancer 2y ago
That's beside the point but you are probably referring to harry potter and the chamber of secrets not the half-blood prince.
48.
Show HN: Voil – Oil.nvim for VSCode(marketplace.visualstudio.com)
5 points by hexomancer 2y ago|1 comments
49.
by hexomancer 2y ago
Page 31 is interesting, where apparently in the task of creating PRs for an internal repository the o3-mini models by far have the lowest performance (even worse than gpt-4o). What is up with that?
50.
Show HN: Edit the filesystem like a text buffer in VSCode(github.com)
2 points by hexomancer 2y ago|0 comments
51.
by hexomancer 2y ago
If I publish some c++ code that has some hard-coded magic values in it, can the code not be considered open source until I also publish how I came up with those magic values?
52.
by hexomancer 2y ago
A- Who cares? B- The latex source of the book is available on the ArXiv page.
53.
Deepseek R1 Zero(huggingface.co)
4 points by hexomancer 2y ago|0 comments
54.
by hexomancer 2y ago
I wouldn't call the case with two sequence "some narrow version". It is by far the most common instance of the problem and honestly the first thing that pops into my mind when I hear LCS, that's why I was very surprised
55.
Ask HN: How much performance can be gained by etching LLM weights into hardware?
3 points by hexomancer 2y ago|2 comments
56.
by hexomancer 2y ago
This works the same in vscode as it does in vim as far as I am aware. e.g. `:vs` to split the window and moving between windows uses the same shortcuts as it does in vim.
57.
by hexomancer 2y ago
All of those are available in the vscode plugin.
58.
by hexomancer 2y ago
Can you give example of vim features that you miss in the vscode plugin? Mainly because I likely don't know about them and I would like to add them to my toolkit (the surround thing is not a vim feature, it is a plugin, although the vs
59.
by hexomancer 2y ago
Nope, with reverse-debugging enabled it will be insanely slow, not something that you can miss. I am curious does your project have large external dependencies or is it self-contained.
60.
by hexomancer 2y ago
Have you actually compared the debug performance of a large cross-platform application on windows vs linux? I am not saying that you haven't just trying to make sure that your argument is backed by data rather than heresay.

More ›