Max pfp

Max

@anakvad

67 Following
93 Followers
Max pfp
Max
@anakvad
How do you solve the quick visual definition of the header level? Just font size isn't enough for me, nor do I want indentation, and I don't want a separate theme for that. So for now, the best I've come up with for myself is to put a unicode number icon on the header level, and highlight it with a color. Just like I did in Emacs.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
If you mean file name, it can be quickly edited via “Rename file” (F2). If you mean headings, it is convenient to switch between them with the “Go To Heading” plugin (on the screenshot). With the help of the QuickAdd plugin you can name files: {{date:YYYY-MM-DD}}. With Templater you can insert the date anywhere: <% tp.date.now(“YYYY-MM-DD”) %>
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I used Terminus for about 10 years, now it's mostly just Jetbrains mono.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I use sed/awk/tr, often with ChatGPT, which will quickly write me an entire pipeline. But even more often I use the excellent search and replace features in Emacs.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I encounter chauvinism from Cloudflare towards Linux users almost every day :). The thing is that on the same laptop, on Windows I don't have to click a checkbox to pass their captcha, but on Linux I always have to do it. It seems Cloudflare's anti-fraud system considers penguins to be very suspicious individuals. But of course, this is very understandable, since desktop Linux users are less than 5%.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Finally was able to connect my wallet, but only through Brave. https://go.moxie.xyz/s/A51
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Same thing, I just found out now and decided to try Talent Protocol. I have tried Metamask & Rabby via Google Chrome & Firefox without success. Now finally installed Brave, which I've been wanting to move to for a long time and everything works fine there. I had the same thing with Moxie.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Decided to start a blog on /paragraph. For now, I will copy posts from my Telegram channel, which I've been keeping for some time, mostly for myself. https://paragraph.xyz/@science-fragments/why-do-we-hiccup
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I often torn (mostly mentally) between Org & Obsidian for my notes, but still go back to Emacs. The two main advantages in Obsidian for me: 1. Having a good mobile app. Yes, there is Orgzly [revived], but I don't like its approach to synchronization. Yes, I can easily run Emacs via F-Droid, but I haven't found any normal keyboard with C & M modifiers. But even with it, apparently it will be inconvenient. 2. Much richer graphical capabilities because of using Electron. It's not critical for me, but sometimes I wish it was. I'd be interested to know who does what with this if experiencing the same problems.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Participating in a bounty from @jake I tried to do a file-tuning of the GPT model on the texts from his blog. Yeah, apparently that wasn't the original idea, but I took the word “train” too seriously, and I didn't want to just give ChatGPT the text and ask him to write an article. Now I'm interested in the task itself, because I'm going to do similar things in the future. Maybe someone can give me some pointers: 1. What did I did wrong and how can I improve the result? 2. What is the best technique to apply for automatic evaluation of the final generation? https://warpcast.com/anakvad/0x6438f3ab
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
No one seems to have tried solving the problem via fain tuning. I tried it, but after spending quite a bit of effort, I seem to have gotten poor results. Nevertheless, here's what I did: https://github.com/keatumal/playing-jake-bounty?tab=readme-ov-file
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Here's what I came up with, but the solution isn't perfect. You use a logarithmic scale, so I added a fake clanker version called “Bar chart trick” which has the number of tokens equal to 0.1 (because a logarithmic scale cannot start from zero). You can adjust this value to emphasize to a greater or lesser extent groups with one token. https://dune.com/queries/4473816/7481940/
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I spent almost 4 hours on this and only realized about the merch after your comment :)
1 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
At first I wanted to do this in an automated way, but the task turned out to be more difficult than I thought. So for the moment I just collected everything from open sources, as I think others have done, so the results will be duplicated. In general, we can use the OpenSea API to get collection descriptions and search there for CC0 traits by keyword or via LLM. The list of collection names to search for can be uploaded from Dune. I can do it, but not for merch :). https://docs.google.com/spreadsheets/d/1BEk-kiqSbHEDG3UE0W1Xn1GCzesGXXOVKzUimH_6Oa4/edit?usp=sharing
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Done
1 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Can you please tell me what exactly needs to be synchronized? Configs, packages, user data, whole system, .... This determines how the task is accomplished: rsync, git, Ansible, ...
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
There's nothing on CMC either.
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
I wrote a script that gets all the tokens through the CoinGecko API. There is not a single token out of 15956 tokens according to this criteria. There is only one token with a ticker greater than 21 characters, but it reached mcap $1.66M in October 2023. Might try other sources. https://www.coingecko.com/en/coins/dev-smashed-his-keyboard
0 reply
0 recast
0 reaction
Max pfp
Max
@anakvad
Created a task that is relevant to me right now. It is a pity that during creation all newline characters are removed in the original description. https://www.bountychain.xyz/bounty/0x38e1d2c61c2803d02f42f6961a3ef08eb9928c631d30296a6ca343056c10be77
1 reply
0 recast
1 reaction
Max pfp
Max
@anakvad
Yes, that's what this method is used for. In SD you can generate an image based on another image - Img2Img. The output image can either be very different from the original image or identical. The point is that you insist on getting an identical image, but ControlNet changes the pose of the character on it.
0 reply
0 recast
0 reaction