Guidance by the United States Patent and Trademark Office, released on Feb 13: Federal Register Money quote: However, a significant contribution could be shown by the way the person constructs the prompt in view of a specific problem to elicit a particular solution from the AI system. (German tech press article: Heise) Read more 15 Feb 2024 - less than 1 minute read
A poster on LinkedIn generalizes from his experience with the Gemini Advanced system to all vision models: Vision models are weirdly prone to prompt injection - they are more likely to take (even contradictory) instructions from an image to follow them. I couldn’t reproduce this: Not reproducible with the Gemini Pro Vision model via Goog... Read more 10 Feb 2024 - less than 1 minute read
Peter Gostev elaborates on LinkedIn about how Google Bard includes 5 hidden chat messages at the beginning of each conversation “to get in the mood”. I found this technique helpful with GPT-4 as well, and have included a small helper tool in my OAI chat for this. My comment on LinkedIn: This can be seen as an example of few-shot prompting. I... Read more 28 Jan 2024 - less than 1 minute read
This work is a gold mine: Mastering LLMs for Complex Classification Tasks. I expect the problems they describe to generalize beyond classification tasks (and has an immediate impact on advanced RAG pipelines). They conclude that: The Claudes have been a somewhat mysterious disappointment, both in their extreme sensitivity to the prompt struct... Read more 19 Jan 2024 - 1 minute read
Particularly to aid with the upcoming events planning season, I have made some substantial updates to my OpenAI- and Anthropic-based chatbots: Export & import of the chat history Word file .docx support File download Support for Claude 2.1 Chat history export Uses for this I can think of: Save/backup your LLM session, and possi... Read more 15 Jan 2024 - 4 minute read