You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have a situation where I have a growing list of tools for GPT/Claude to use; 20 now, and likely 50-100 in the near future. Many tools have 5-10 params each with detailed documentation. With normal tool selection mode, all tools are fully in context for all prompts. Besides for being expensive, I suspect this affects competence because of all the extraneous tools (although I have not evaluated this).
Has anyone tried an approach involving breaking tool use into two steps, i.e. 1) tool selection, and 2) selected tool use? First, you provide just the names and descriptions of all available tools, and output the name of the selected tool. Then in the second step, give it the full specification (including parameters) of the selected tool and have it configure / specify the parameters. Then execute the tool.
I am wondering if anyone has tried this strategy, and found if it's a useful way to scale tools, or found if it's overcomplicated and less useful. Or in general guidance on how to deal with having many tools.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
I have a situation where I have a growing list of tools for GPT/Claude to use; 20 now, and likely 50-100 in the near future. Many tools have 5-10 params each with detailed documentation. With normal tool selection mode, all tools are fully in context for all prompts. Besides for being expensive, I suspect this affects competence because of all the extraneous tools (although I have not evaluated this).
Has anyone tried an approach involving breaking tool use into two steps, i.e. 1) tool selection, and 2) selected tool use? First, you provide just the names and descriptions of all available tools, and output the name of the selected tool. Then in the second step, give it the full specification (including parameters) of the selected tool and have it configure / specify the parameters. Then execute the tool.
I am wondering if anyone has tried this strategy, and found if it's a useful way to scale tools, or found if it's overcomplicated and less useful. Or in general guidance on how to deal with having many tools.
Beta Was this translation helpful? Give feedback.
All reactions