Just an FYI that the latest Git Pull has dramatically decreased the accuracy of data coming from Gemma. Where the LLM was previously unable to provide data the Google search augmented search is pulling wrong information. An example is there are apparently 9 seasons of The Rookie a tv show which is wrong. Also when asked how many medals New Zealand win at the 2024 Paris Olympics, she said 1.
I'm unclear at this stage if the removal of common words, which is new code is contributing to this or if it's the Google search results.
Great to see so much development going on with this project!
Hey Team,
Just an FYI that the latest Git Pull has dramatically decreased the accuracy of data coming from Gemma. Where the LLM was previously unable to provide data the Google search augmented search is pulling wrong information. An example is there are apparently 9 seasons of The Rookie a tv show which is wrong. Also when asked how many medals New Zealand win at the 2024 Paris Olympics, she said 1.
I'm unclear at this stage if the removal of common words, which is new code is contributing to this or if it's the Google search results.
Great to see so much development going on with this project!
<!-- github-sync-id:2507423103 -->
without google results the LLM couldn't accurately say what 2024 they dont have up to date info, the LLM are not perfect at all. BUT if you liked the old version you can still go back easily try changing the line in llm.py https://github.com/SpudGunMan/meshing-around/blob/7afd6bbbe91e17f718ca51a8e3b7e1cdb7fcd19a/modules/llm.py#L14
change this to False
<!-- github-comment-id:2332115367 -->
Thanks, I've implemented this and it seems better! - Another question if I may please. The LLM even with Google disabled doesn't seem to work without internet connectivity. Is this intentional? - I'd like to operate this in a Meshtastic Only environment where the R-Pi is offline.
Thanks, I've implemented this and it seems better! - Another question if I may please. The LLM even with Google disabled doesn't seem to work without internet connectivity. Is this intentional? - I'd like to operate this in a Meshtastic Only environment where the R-Pi is offline.
<!-- github-comment-id:2333139474 -->
cleaned this up a lot, it was messy! Yes it should work fine offline.
fixed in https://github.com/SpudGunMan/meshing-around/commit/819c37bdec401eeb7363a8ba84da96af02c4ea04
<!-- github-comment-id:2333311788 -->
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Hey Team,
Just an FYI that the latest Git Pull has dramatically decreased the accuracy of data coming from Gemma. Where the LLM was previously unable to provide data the Google search augmented search is pulling wrong information. An example is there are apparently 9 seasons of The Rookie a tv show which is wrong. Also when asked how many medals New Zealand win at the 2024 Paris Olympics, she said 1.
I'm unclear at this stage if the removal of common words, which is new code is contributing to this or if it's the Google search results.
Great to see so much development going on with this project!
without google results the LLM couldn't accurately say what 2024 they dont have up to date info, the LLM are not perfect at all. BUT if you liked the old version you can still go back easily try changing the line in llm.py https://github.com/SpudGunMan/meshing-around/blob/7afd6bbbe91e17f718ca51a8e3b7e1cdb7fcd19a/modules/llm.py#L14
change this to False
Thanks, I've implemented this and it seems better! - Another question if I may please. The LLM even with Google disabled doesn't seem to work without internet connectivity. Is this intentional? - I'd like to operate this in a Meshtastic Only environment where the R-Pi is offline.
cleaned this up a lot, it was messy! Yes it should work fine offline.
fixed in https://github.com/SpudGunMan/meshing-around/commit/819c37bdec401eeb7363a8ba84da96af02c4ea04