Your AI Remembered. Did It Learn the Right Lesson?For Trip A, you give an AI travel assistant one restriction: For this trip, do not suggest flights before 10:00. It has two options: Departure Price 08:00 $180 13:00 $240 These fares and t11G3d ago 路 9 min readanandms101.hashnode.dev
An Attention Window Is Not a Checkpoint: Inside vLLM's 128-Token ReplayYou ask a coding agent to fix a retry bug: Fix the failing retry test. Keep the public API unchanged. It reads the client code, opens the test, makes a small edit and runs it. The test still fails. 12KAOct 11 路 10 min readanandms101.hashnode.dev
The Model Is Untrusted Code: A Design Review of Meta's MuseYou ask a personal agent to prepare notes for tomorrow's meetings. It reads your calendar, pulls the related email threads, and opens your project notes. One thread came from outside your organization1MSep 29 路 17 min readanandms101.hashnode.dev
Toon vs JSON: The Battle for Token Efficiency in LLM ApplicationsLast month, I was reviewing our API costs and noticed something alarming: we were spending way more on LLM tokens than expected. Every API call to GPT-4 or Claude was eating into our budget, and I couldn't figure out why until I started looking at wh...Nov 8, 2025 路 6 min readanandms101.hashnode.dev
From 0 to On-Chain: My First Web3 dApp for Tracking In-Game Taxes馃З Introduction: My First Web3 Hackathon Until a few weeks ago, I had only scratched the surface of blockchain, the usual stuff: watched a few videos, tried out MetaMask, maybe deployed a test token once. But during the recent Polkadot 脳 EasyA Hackat...Apr 30, 2025 路 9 min readanandms101.hashnode.dev
