Fixed search prompt

[portoaj]

Currently the search prompt reduces the creativity of the LLM to a detrimental degree to the point that it can't search for anything that isn't presented in an extremely simple input query. In particular, if I try to specify the schema or the answer I want in my prompt the search that is output is unhelpful. Here's an example of a query I'd like to run that fails currently:

graph_config = {
    "llm": {

        "model": "gpt-4o-mini",
    },
    "verbose": True,
    "headless": False,
    "max_results": 5
}

class AnswerModel(BaseModel):
    answer: Literal['yes', 'no']


search_graph = SearchGraph(
    prompt="Does the Alienware Aurora R15 Desktop have bluetooth support? Your output should be of the form: {\"answer\": \"answer here\"}. The answer should be exactly one of these strings: 'yes', 'no'",
    config=graph_config,
    schema=AnswerModel,
)

Here's the search term that gets output currently: Search Query: {"answer": "yes"}
Here's the search term that gets output after my change: Search Query: "Alienware Aurora R15 Desktop bluetooth support"

This simple prompt change should make the search internet node much more useful. In my example, it goes from a completely useless query to the exact query I would've expected.

Please let me know if there are any other changes you think I should make to the prompt to make sure that it doesn't break other people's use cases.

[portoaj]

Realized I was merging from/into main instead of pre/beta like the other PRs

[f-aguzzi]

Thanks for your contribution. I moved your new prompt to a different file, so now everything is ready for merging.

[github-actions]

🎉 This issue has been resolved in version 1.14.0-beta.12 🎉

The release is available on:

• v1.14.0-beta.12
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This issue has been resolved in version 1.14.0 🎉

The release is available on:

• v1.14.0
• GitHub release

Your semantic-release bot 📦🚀