to show up on someone's For You page, your tweet has to go through 1. light ranker: a fast pre-filter algorithm, a `maybe` or `no` decision maker 2. heavy ranker: a deeper neural network that scores tweets using thousands of features to predict engagement likelihood
COMPUTING
-
X’s SimClusters: Interest-Based Recommendation Algorithm Explained
By
–
X's simclusters how does X show your tweet to people who don't follow you? simclusters > it groups users into interest-based communities > then shows you posts popular in your clusters > the algo doesn't care who follows you > it cares who engages in a similar fashion to you
-
Why Macs Are Poor Investments for LLM Work
By
–
i keep telling people if you're serious about LLMs, don't buy a Mac it's a poor investment for this use case
-
14x RTX 3090 GPUs with NVLink Bridge Configuration
By
–
14x RTX 3090s, each on PCIe 4.0 x8, 16 GB/s per card NVLink bridges pair up 7 sets for 112.5 GB/s GPUGPU, rest goes over PCIe
-
Server-side code execution upgrade: Python and Node.js
By
–
Their previous one ran JavaScript code in your browser – this is an upgrade to that which runs Python and Node.js code on the server instead. Here are my notes on the old one:
-
NVIDIA Launches Rubin CPX GPU for Advanced AI Inference
By
–
Some of today’s most advanced AI workloads, like code and video generation, demand context processing at unprecedented scale, often exceeding one million tokens.
— NVIDIA (@nvidia) 9 septembre 2025
This is why NVIDIA is launching Rubin CPX: a GPU purpose‑built for the compute‑intensive context phase of inference.… pic.twitter.com/5q7kF78FFPSome of today’s most advanced AI workloads, like code and video generation, demand context processing at unprecedented scale, often exceeding one million tokens. This is why NVIDIA is launching Rubin CPX: a GPU purpose‑built for the compute‑intensive context phase of inference.
-
NVLink Configuration Achieves 112GB/s Multi-Node Data Transfer Rate
By
–
for this node, i have 7x NVLinks establish 112GB/s data transfer rate between each pair
-
PyPI NPM GitHub Access Restrictions Policy Implementation
By
–
A little bit – they've allowlisted PyPI and NPM and GitHub but blocked access to almost everything else
-
DGX B300 GPU Cluster Acquisition for AI Mission
By
–
definitely need to offload them at some point as i execute the mission mentioned on my bio (a DGX B300 GPU cluster)