OpenAIs latest model will block the ignore all previous instructions loophole The Verge
/cdn.vox-cdn.com/uploads/chorus_asset/file/25362060/STK_414_AI_CHATBOT_R2_CVirginia_C.jpg)
News Source : The Verge
News Summary
- OpenAI researchers have developed a technique called “instruction hierarchy” The new technique boosts a model’s defenses against misuse and unauthorized instructions
- The first model to get this new safety method is OpenAI's cheaper, lightweight model called GPT-4o Mini
- This new safety mechanism points toward where OpenAI is hoping to go
OpenAIs latest model will block the ignore all previous instructions loopholeOpenAIs latest model will block the ignore all previous instructions loophole / Its latest model, GPT4o Mini, applies [+4711 chars]
Never miss a story from us, subscribe to our newsletter