Conversational prompt hack

[KenjiPcx]

Issue details

Right now I'm sure you are aware that if you speak something wrongly, you repeat yourself to attempt to tell the agent what you really intended to say

We can actually do a prompt hack here, change the student prompt to make it aware of this, by including something like:

• "You receive user transcripts which might be noisy"
• "If a user repeats themselves, it is likely the later sentence is more correct and they are trying to fix transcription mistakes"

Whoever picks this up, please run tests to make sure everything works