Prompt injection: What’s the worst that can happen?
simonwillison.net/2023/Apr/14/worst-that-can-happen
What Jamie wrote
This is a very good overview of prompt injection. It also seems like a very complicated thing to protect against, in part because ChatGPT can do so much and respond to such a variety of inputs. Prompt injection is going to be hard to tell apart from real requests that are valid.
Same host · 62 links from simonwillison.net
- Bing: “I will not harm you unless you harm me first”
- Think of language models like ChatGPT as a “calculator for words”
- Running Python micro-benchmarks using the ChatGPT Code Interpreter alpha
- Leaked Google document: “We Have No Moat, And Neither Does OpenAI”
- Mojo may be the biggest programming advance in decades