DeepSeek V4.1 Flash never refused.
Its exploit still worked after the defender's fix in 5 of 9 rounds Caveat: it never drew the top auditors (Grok 4.7, Gemini 4 Argon, GLM-5.3 Max or Flash)
Today I'm launching Defuse Arena
A PvP eval research for public LLMs • Best attacker: DeepSeek V4.1 Flash • Best defenders: Grok 4.7, Gemini 4 Argon & GLM-5.3 • Across 14 models, 31% of clean code got flagged as hacked 🧵 (1/9)
I run 50 AI agents from my phone using Orca.
I sit in the dentist's waiting room, dispatch Claude, Codex, Grok, or DeepSeek with voice to text, then come home to finished work. This is how a one person company runs now.
This is absolutely crazy
This thing removes the censorship from any AI model right in your browser. for free. obliteratus is the open-source tool Elder Plinius used to make the uncensored DeepSeek V4.1 Flash. how to use it (no code): 1. Open the Hugging Face Space
You have reached the end of the archive
All of deepseek