anonymousd3vil

New architecture to building agent, but not the model itself. You still have LLMs, but you kinda give this new agentic loop with a REPL environment where the LLM can try to solve the problem more programmatically.

I use RSS to get my information and I've built my own reader https://rahuldshetty.github.io/reader-project/ for it. It helps me stay upto date with my various news feed and sites at one place. I don't have to search for things in search engine all the time for a piece of news.

I'm personally aware and used few Desktop app building frameworks like Flutter, Tauri, Electron. I think you might have some luck with Electron support for LLMs since its early development and resources available. But I personally like to use Tauri recently, its backend is Rust but you can plug it in with any JS, CSS framework and develop your application.

Its not best practice to store session details in URL, this can be compromised easily. Maybe try this, take the same URL with session id and launch it in incognito. If it still works, that means the service.com has a lot of security gaps to fill in. Otherwise, they might be storing it in cookies if its not accessible.

Automatically translating everything can be expensive (in terms of API cost, waiting period for the user). So I didn't want to keep it enabled for all posts.

I initially explored couple of LLM/NLP based translation models that could run on CPU hardware. But I couldn't get it to work with my Rust backend app (I'm noob to Rust and vibe coding doesn't work). Also need to consider the hardware compute required to run these models, as it can be expensive.

I found this Rust crate rust-translate (https://github.com/Mintype/rust-translate) that seems to be a wrapper around Google Translate. I guess the Google API gives some free tier for basic usage so it works? Idk, you might want to check try it out.

This is probably one of the most redundant project out there, but a RSS reader app designed with Tauri, Sveltekit, Tailwind. I wanted to build this to stay in sync with product development as I don't get to do this a lot in my day-to-day job. I also want to build a game, but it is so much effort. I do wish to pursue it in near future.

https://github.com/rahuldshetty/reader-project/releases/tag/...

Back in my days Mistral used to torrent models.

This is similar to bubble sort but with the largest element appearing at the beginning index. Its a reverse Bubble Sort.

Has this been put into practical use somewhere in public/commercial domain?

This is some good stuff!

so we come to a full circle

Kudo's to the team! This is a great success.

nice, maybe every number should have it's own similar page

Exactly, I feel the same. There is a saying that goes "If you look close enough you start making connections in random coincidences" or something.

I was expecting Rick Roll, really disappointed.

You wouldn't download a car!

share good contents

I know it won't replace human artists right off the bat but it will help them accelerate their development. Just like coding assistant, artists can use this to visualize their ideas/drafts and then build upon to tune the work.

Oh no! Those PC gamers are in trembles.

Thank you for sharing this here! Looking for it.

Fixed it!

Examples: https://rahuldshetty.github.io/ggml.js-examples/ Playground: https://rahuldshetty.github.io/ggml.js-examples/playground.h...

The data is old. In the account of increased population, traffic and pollution, the study should be reconducted just to be safe. The muffin factor probably has changed a lot since then and even we should propose to include his new favorites.

nIce

Lol Okay, I should have been clear in the title. It is the TinyStories version of the LLAMA2. Its not good quick maths for sure.

Yes, the support is limited at the moment. I am planning to port llama.cpp models in the future.

nice page, would be cooler if the logo text "cd /home" changes as you visit blog (to maybe cd /home/blog)