Gregory's Server@alyssa what are your thoughts on isolated decision making behind a fixed input/output layer? For example imagine if Siri etc. Used a LLM to interpret and do language parsing but NOT to build the response
|
4 comments
@erisceleste I mean, isn't one of the most common issues with virtual assistants that they misinterpret a query? For example, consider the following exchange with ChatGPT that Siri (and probably others??) would've completely failed at. Note that it properly parsed out the request AND corrected my spelling.  perhaps i speak from too much of a cultural bubble when i say that groceries:append(apples, 1400) seems better to me than using english to describe a task that doesn't need to be described in english, in the first place this can be three button-presses and a swipe of a drop-down @erisceleste I appreciate why you would feel that way, but I hope you understand that the general population just wants to be able to make a request and have it understood, not following a syntax |
@ezekiel
cool but you could make it better by getting rid of the bit where it does language parsing
(serious answer, i cannot think of a justification to myself for why this is an improvement over just _not_ having that step)