Do the thumbs up in prod affect LLM understanding

[sofiahd]

Hi all,

my team is currently testing the new patch 6 for NowAssist. Thumbs up/down seem to be training the LLM in Dev2.  Something I'm wondering though is if the training thumbs up/down in prod will affect LLM training also meaning that if users are not clear on procedures or dislike a procedures may affect 'the truth' of the LLM.

potentially is that something that can be kept for dev2 for training purposes and have thumbs up/down show in prod but only as feedback that don't affect LLM training?

[anubhavkapoor76]

@sofiahd - Feedbacks are logged in table sys_gen_ai_log_metadata

Also thumbs up/down are for software engineering teams to understand how AI model is performing and tune it for future releases. Its not used for model retraining or fine-tuning.

[Stefan Budovsky]

Hello Anu, could you please clarify? Does this mean that if you ask the same or a very similar question, you will continue to receive a very similar answer, regardless of how many thumbs down it receives? Thank you.

[Stefan Budovsky]

Hello Anu, could you please clarify? Does this mean that if you ask the same or a very similar question, you will continue to receive a very similar answer, regardless of how many thumbs down it receives?

[anubhavkapoor76]

@Stefan Budovsky - The responses could be different each time (as LLM's are probabilistic), however the thumbs down/up does not affect the responses.

@sofiahd - Request you to please close the thread and help other community members.