Good to be reminded that many are still just using the chat interface for quick answers!
Like the framing, will definitely use a variant in my work. We experimental enthusiasts help others most by sharing concrete examples of prompts and outcomes across the different layers, not just 'showing-off' final results. The interesting part is what you put in, when, and how you correct during the full process.
The branching tree is how I work currently: deliberately forking to create different results and expand possibilities. Exploring alternatives relatively fast and cheap, that for me is the true current value of AI. The danger is that you keep on exploring for ever..
Prompting techniques to push AI outside standard answers is something we need to spread more. Just came across this Harvard article you may like: https://gking.harvard.edu/quest
I like the layered model. Where in the layers does a "harness" fit in? It seems increasingly clear that using a harness on top of an AI model that knows how to manage context, how to bring in the right skills at the right time, how to support particular activities through invoking sub-agents, etc. can get dramatically better performance than if the same prompt was presented to the AI model directly.
Good to be reminded that many are still just using the chat interface for quick answers!
Like the framing, will definitely use a variant in my work. We experimental enthusiasts help others most by sharing concrete examples of prompts and outcomes across the different layers, not just 'showing-off' final results. The interesting part is what you put in, when, and how you correct during the full process.
The branching tree is how I work currently: deliberately forking to create different results and expand possibilities. Exploring alternatives relatively fast and cheap, that for me is the true current value of AI. The danger is that you keep on exploring for ever..
Prompting techniques to push AI outside standard answers is something we need to spread more. Just came across this Harvard article you may like: https://gking.harvard.edu/quest
This is one of the most useful explanations I’ve read Sami. Thank you 🙏
I like the layered model. Where in the layers does a "harness" fit in? It seems increasingly clear that using a harness on top of an AI model that knows how to manage context, how to bring in the right skills at the right time, how to support particular activities through invoking sub-agents, etc. can get dramatically better performance than if the same prompt was presented to the AI model directly.