ThePrimeagen: closed model gardens won't stop attackers already using abliterated models
max_paperclips · x · 2026-10-23
Developer ThePrimeagen argues that hiding models behind closed doors while attackers already use abliterated (uncensored) models was never a winning strategy. He suspects the封闭 approach is almost intentional and predicts "weeks of chaos" as everything gets hacked — a pointed entry in the open-vs-closed AI security debate.
Related event: $10 Uncensored GLM-5.3 Sparks 'Vibe Hacking' Security Debate(8 posts)→
More from Fun
- Economist Jason Furman vibecoded a prototype in one night, now sits on 500k lines of code — soumitrashukla9 · 2026-10-24
- Fable 5.5 taking so long it 'probably escaped the sandbox' — prasenx · 2026-10-24
- 'This is my quant' gets roasted: can the model solve 24 from 3, 3, 8, 8? — markchen90 · 2026-10-24
- NativeCAD generates a full A320neo from prompts alone, flyable with a working cockpit — burhop · 2026-10-24
- Feeling blue: Klein-blue brushstroke aesthetic images go viral in AI circles — gizakdag · 2026-10-24
- AI meme mashes up Jarvis, Ultron and HAL 9000's classic refusal — generativist · 2026-10-24