surya on Suriya's site

I reused the functions provided at https://github.com/jepsen-io/maelstrom/tree/main/demo/go for interfaces. repo: https://github.com/subygan/gg 1. echo This was a gentle intro to maelstrom and it’s tools.

This is a (mostly) search engine specific requirement where individuals are protected from the harm of their past actions. disparity in countries Similar to any privacy issue, these countries are divided in the way these laws are implemented.

My aim for this blog was to keep it as not too technical as possible.

number of lines in a file wc -l file.txt split file into multiple files split -l 1000 file.txt append two files into one cat file1.txt file2.txt > file3.txt find pid of port sudo lsof -i :<portnumber>

Failing to read the documentation and understanding the code underneath is the bane of all software bugs and I’ve guilty of it on multiple counts

privacy problems and attacks in machine learning Membership inference attacks: These aim to determine if a particular data point was used to train a model.

I used to deploy everything in an ec2 instance with docker-compose and nginx.

I experimented with a lot of models, and as advertised claude3-haiku gave the best results, in terms of quality and cost.

opinions i’ll defend, bets i’m making, and things i was wrong about. argue with me at suriyaganesh097 at gmail.com. controversial opinions i have pessimism is defaulting to laziness. optimism is the harder, rarer skill. you should have enemies. say “no” to

I like to have a standard work environment that I can have multiple zellij is important, installing it with, cargo install zellij

global gitignore I’m surprised I only came to know on 2024 that you can have a global gitignore you can just put this inside ~/.gitignore

A simple way to include all available libraries in cpp in most OSs is to just add #include <bits/stdc++.h>.

exercise 1.1.1 Counter example: {0,0},{0,-1},{0,-2},{0,4},{0,-4},{4,4} Greedy value: 6 + 1 + 4 = 11 Correct value: 4 + 4 + 1 = 9

The growth of Paul into a sort of Genghis Khan style oppressive ruler was interesting.

Dune was such a surprising read for me.

I’ve been getting into addicted to Coffee since last year.

My friend Limin and I, were playing around with some LLM agents and realized that these agents could be a very good way to learn a language as they have a much better grip on the language and are able to provide great feedbacks.

repo Nemo Speaker diarization training MultiSpeaker Simulation data sets LibriSpeech VoxConverse resources Whisper Streaming Diarist

material Usually reading translated text is a little confusing, because I feel like I’m hearing the translator’s voice instead of the author’s voice, or in this case Plato.

Reading input from users is difficult.

This is going to be binary analysis with LLM type thing The idea is to see, if LLMs are able to understand the “language” of binaries and are able to retrieve the underlying vulnerability.

Flutter comes with super complex build system that ties in all the language runtimes with dart and the native rendering engine.

let x; // declare "x" x = 42; // assign 42 to "x", type is inferred or let x = 42 //assigns variable let x: i32; //declaring variable "x" of 32 bit integer foobar(x); //uninitialised variable assignment x = 52; // assigning 52 foobar(x); //x will…

pandas

amazon linux Does not use apt-get and only uses yum. replace all apt-get with yum sudo systemctl enable docker.service sudo systemctl start docker.service

When a column has NaN it is shown as float value in a column. need to use .fillna("") to replace the Nan with empty string.

I’ll add the ones with the most fun:utilization first building graphs using llms: This strategy comes clearly under the by LLMs category and the most obvious one as well.

aiaac, AI incidents database.

The base premise is that, aligning the model for the goal that it was created for is a long tail problem and requires multiple guardrails to control it’s output.

Continual learning ai guardrail comparison There are multipel LLM guardrails, evaluating them would be something privacy detector dataset and custom model There is no dataset to test a model for it’s privacy Capabilities A custom model implementation that…

தோத்திரப் பாடல்கள் - யோக சித்தி தேடிச் சோறுநிதந் தின்று -- பல சின்னஞ் சிறுகதைகள் பேசி -- மனம் வாடித் துன்பமிக உழன்று -- பிறர் வாடப் பலசெயல்கள் செய்து -- நரை கூடிக் கிழப்பருவ மெய்தி -- கொடுங் கூற்றுக் கிரையெனப் பின்மாயும் -- பல வேடிக்கை மனிதரைப் போலே…

இன்றிளைப் பாறுவம் என்றிருந்தால் – வழி என்னென்ன வாகுமோ ஓரிரவில் சென்றிளைப் பாறுக முற்றிடத்தே – தம்பி தேன்வந்து பாயும் உன் நெஞ்சிடத்தே!

உயிர் வருக்கம் 1.

in context learning chain-of-thought Broadly three classes of reasoning tasks “let’s think step by step” seems to be the key phrase

SQUARE (Software Quality Requirements Engineering) is SEI’s framework for “Improving identification, analysis, specification and management (of software)”.

These are my notes from Nishant’s excellent book, Data Privacy: Runbook for Engineers 1 privacy engineering Data Security is usually tightly coupled with privacy.

Resume [pdf] Email: suriyaganesh097 at gmail.com linkedin.com/in/suriya-ganesh github.com/subygan google scholar orcid employment Software Engineer Oxen.ai May 2025 - present VCS system for large datasets using rust.

PurpleLlama is a new framework proposed by Meta to evaluate LLMs for their secure coding ability. cyberseceval This is the core part of the research paper which has created a benchmark for LLMs to test generated code.

Membership inference Attack is a way to retrieve back the probability of the original training inputs. notes https://arxiv.org/pdf/1610.05820.pdf

Suppose an array of length n sorted in ascending order is rotated between 1 and n times.

problem statement Given n non-negative integers representing an elevation map where the width of each bar is 1, compute how much water it can trap after raining.

problem link: https://leetcode.com/problems/longest-valid-parentheses/ Typically, in the parentheses problem a stack is the obvious choice.

My notes on Karpathy’s Intro to LLM talk llama-2-70b comes with run.c ~500 lines of c for it to be a self-contained runnable model.

parkinson’s law “work expands so as to fill the time available for its completion” When Northcote made this observation, this was designed for organisations, but I believe this is a good representation for personal tasks as well.

how are ml systems measured.

Although a translation, this was a good book.

I’ve previously read Jeffrey Archer’s other Short story collection, Twelve Red Herrings and The Collected Short Stories.

Fundamental problem has been that, privacy is hard to define.

The problem is, uncertainty.