Prompting And Prompt Engineering Facing Notable Changes Due To OpenAI Latest o1 Generative AI Model
(The old hidden constraints in new clothing. The user may go in assuming that the app thought of everything only to find out they hit a trap. This flips them to, if it has not been explicitly stated, assume it needs definition. Or a lot of testing. In this case, asking it to be verbose is considered to be against the terms of service since the vendor does not want others cloning models based on its reasoning techniques.)
Google releases ‘prompting guide’ with tips for Gemini in Workspace
Coming in at 45 pages, there are example personas and prompts that go through refinements for: Customer service, Executives and entrepreneurs, Human resources, Marketing , Project management, Sales. Ultimately, Google says to review outputs for “clarity, relevance, and accuracy” before using it.
download the guide here
Anil, C., Durmus, E., Sharma, M., Benton, J., Kundu, S., Batson, J., ... & Duvenaud, D. (2024). Many-shot Jailbreaking.
Long contexts represent a new front in the struggle to control LLMs. We explored a family of attacks that are newly feasible due to longer context lengths, as well as candidate mitigations. We found that the effectiveness of attacks, and of in-context learning more generally, could be characterized by simple power laws. This provides a richer source of feedback for mitigating long-context attacks than the standard approach of measuring frequency of success