I got an LLM to dynamically set its own temperature!
Just wrote a blog post on how I did this and why dynamic temperature adjustment might be a valuable ability for a language model to possess.
Read here: amanvir.com/blog/getting-an-llm-to-set-its-own-temperature
Thanks Xavier, really glad you like the site! 😄
If you’d like to check out some obscure islands that I find interesting... :)
Feels surreal seeing a creative project of mine in such a major publication!
My work was featured in The New Yorker this week!
The "Goings On" section, which focuses on curating cultural touchstones and happenings, linked to my interactive webpage on Earth's obscure islands as part of their annual "Summer Culture Preview."
If you'd like to run the visualization and play around with it, please feel free to check it out on my website: amanvir.com/gpt-2-attention!
I created a 3D visualization that shows *every* attention weight matrix within GPT-2 as the model generates tokens!
I ended up writing an entire article on my website that explains what Dawkins' weasel is and how it can be used to create the cool effect you see in the video below: amanvir.com/blog/animati...
Feel free to check it out, and let me know what you think!
Over the past few days, I worked on creating a text animation effect using Dawkins' weasel, a computer simulation devised by evolutionary biologist Richard Dawkins in the 1980s.
Haha, that sounds awesome! Truly a win-win 😄
If you want to check out the post, please visit: amanvir.com/blog/turning-my-esp32-into-a-dns-sinkhole
I built an ESP32-based DNS sinkhole that stops you from doomscrolling!
This was a really fun project that I wrote a detailed blog post about.
In it, I document my experience building this thing, and I also share what I’ve learned about DNS, networking, and low-level programming along the way ⬇️
Currently, this is a pretty simple program, but I hope that I (or anyone else in the community!) can expand on it by adding functionality to tackle typos, gibberish text, and other issues found within public domain literature.
Here’s the GitHub repo for this project:
github.com/amanvirparha...
These indices are used to create a new ePub that *only* includes meaningful parts of the book's text - the parts that we're interested in reading!
The model is asked to respond with JSON that contains the indices for the first and last paragraphs of each chapter. Gemini is able to accurately return the index of any given paragraph, because each “paragraph” from the book’s raw text is wrapped with a numbered tag which the model can reference:
The script I ended up writing works like this: the model first takes in a prompt and the book’s content. It's not asked to return a corrected version of the entire book (output limit = ~8k toks), but is instead asked to return JSON that aligns with a provided schema (structured generation).
The model also performs pretty well at “needle In a haystack”-type tests, so it seemed like the perfect choice! (some tweets by @jeffdean.bsky.social!)
I started thinking about fixing some of these typesetting/formatting issues using an LLM, and I quickly realized that Gemini would be perfect for this project - the 1M+ token context window would allow me to paste the text for an entire book in one shot.
In my experience, most books on Project Gutenberg are unfortunately formatted in a non-standard way, and a lot of books are actually plagued with much bigger issues than the one I was looking for.
This is essentially the reason why initiatives like standardebooks.com exist.
The ePub I found wasn’t properly formatted: it contained a duplicate, malformed copy of the table of contents spanning several pages, and the worst part was that every single poem was displayed in monospace for some reason.
This project was born out of a personal need: I was looking for a poetry collection by Robert Frost on gutenberg.org, a repository with thousands of free eBooks, but when I downloaded the ePub for this particular book, I was rather disappointed.
I wrote a script that uses Google’s Gemini-1.5-Flash model to fix typesetting and formatting issues in public domain eBooks from Project Gutenberg.
It’s open source, and it works thanks to Gemini’s insanely long context window (1M+ tokens)!
A thread 🧵 ⬇️
Writing a dithering algorithm in Racket
amanvir.com/blog/writing...
#lisp #scheme #racket
Hey, I'm the creator of this site!
Thank you so much for your kind words, I really appreciate it!
Spending time going through this gallery of "interesting islands" makes me very happy. Like poring over the Brittanica atlas as a kid and reading stories of the bermuda triangle, read to me by Jed Barlett.
amanvir.com/obscure-isla...
Thanks for posting my blog here on Bluesky!
I hope you enjoyed reading it :)
Nice site of fun facts about obscure islands:
amanvir.com/obscure-isla...
(Via Hacker News)
Thanks for sharing this, Jeff!
I made this site a little over a year ago as part of my series of publicly documented "internet experiments" (amanvir.com/experiments), so I'm really glad to see it getting some love :)
Hey Jeff, I'm the creator of this visualization :)
The sense of childlike wonder you describe is the exact reason why I chose to build this project!
I remember spending hours staring at my globe as a kid, learning about new and unfamiliar places. Truly a foundational childhood memory!
