That approach does a lot better, but LLMs still have positional bias problem baked into the transformer architecture (https://arxiv.org/html/2406.07791v1 https:

That approach does a lot better, but LLMs still have positional bias problem baked into the transformer architecture (https://arxiv.org/html/2406.07791v1 https://arxiv.org/html/2406.07791v1). This is where the LLM biases selecting information earlier in the prompt than later, which is unfortunate for tool selection accuracy. Since 2 steps are required anyways, might as well use a dedicated semantic search for tools like in agents.json.