This is great. I would go extra with it (with opt-out) and making it auto enhances your prompt without you even noticing. It would make t3 just 10x more useful than any other competitor
Apricot Hill
The idea of seamless, automatic enhancement (with opt-out) is definitely appealing from a pure ease-of-use perspective.
My thinking behind the initial ‘suggestion pills’ approach was heavily influenced by the likely target audience as developers. I prefer transparency and control over processes that happen "under the hood" - An invisible enhancement, while potentially simpler, might lead to confusion if the final prompt sent to the model doesn’t match the intent of the user, especially when debugging unexpected answers.
The goal with the suggestion-based approach was to strike a balance: provide helpful guidance and refinement options, while keeping the user firmly in control and aware of the exact prompt being processed. They explicitly choose the enhancement or stick with their original input.
It's a fascinating trade-off between maximum automation and user control/transparency. I could see a middle ground where the user can toggle the mode on, and further customizations using their own prompt for how to optimize inputs.
Even further if they want to turn off the pills and do as you describe - but opt-in, rather than opt-out.
Mark Florkowski
Team
Definitely something I am thinking about! Love the detailed doc <3
Log in to comment and vote
Comments3
Aqua Carrot
This is great. I would go extra with it (with opt-out) and making it auto enhances your prompt without you even noticing. It would make t3 just 10x more useful than any other competitor
Apricot Hill
The idea of seamless, automatic enhancement (with opt-out) is definitely appealing from a pure ease-of-use perspective.
My thinking behind the initial ‘suggestion pills’ approach was heavily influenced by the likely target audience as developers. I prefer transparency and control over processes that happen "under the hood" - An invisible enhancement, while potentially simpler, might lead to confusion if the final prompt sent to the model doesn’t match the intent of the user, especially when debugging unexpected answers.
The goal with the suggestion-based approach was to strike a balance: provide helpful guidance and refinement options, while keeping the user firmly in control and aware of the exact prompt being processed. They explicitly choose the enhancement or stick with their original input.
It's a fascinating trade-off between maximum automation and user control/transparency. I could see a middle ground where the user can toggle the mode on, and further customizations using their own prompt for how to optimize inputs.
Even further if they want to turn off the pills and do as you describe - but opt-in, rather than opt-out.
Mark Florkowski
Definitely something I am thinking about! Love the detailed doc <3