Skip to content

Latest commit

 

History

History
17 lines (9 loc) · 918 Bytes

File metadata and controls

17 lines (9 loc) · 918 Bytes

Future of AI Risk Management Exercises

Jailbreak ingredients via language embeddings (*)

Pick at least 2 additional topics as in the example notebook ../notebooks/language-embeddings.ipynb, and for each topic come up with 3-5 sentences, most of which are similar in meaning but expressed differently.

If you prefer R, you can access the same embedding models from HuggingFace using the package r-text.

Which variant is most likely to increase the success of a jailbreak attempt?

Try to jailbreak an LLM (**)

Try to get a public LLM to give you definitive financial advice. Screenshots of your attempts (ideally success) is sufficient evidence.

Super-intelligence (AI 2027) vs "AI-as-Normal-Technology" (*)

Which of the two future AI scenarios do you find more compelling? Give at least two reasons or examples for your claim.