My thoughts
Commercial models refuse a lot of legitimate requests. Here's why "uncensoring" an open-weight model is useful and how I achieved it. ⚠️ A note before we start. An uncensored model will answer almost...
Purple used to mean the void, the threshold, the half-demon. Then a single Tailwind decision in 2020 propagated into every AI's training data, and now it means a generic SaaS landing page. A meditation on what AI does to cultural signifiers.
One Month Into Local LLMs: What I've Learned It's been about a month since I went down the local-LLM rabbit hole. In that time I've: Built llama.cpp from source Written my own benchmarking software to...
Andrej Karpathy's hour-long LLM talk is from late 2023. Most of it still holds up. A reread two-and-a-half years on.
Two techniques worth remembering from TryHackMe's Opacity — bypassing PHP upload filters with a fragment character, and cracking .kdbx files with keepass2john + john.
How an exposed /actuator/sessions endpoint on the HTB CozyHosting box turns into a full authentication bypass — and a refresher on the sudo ssh GTFObin.