the LLM makers market their product for quick answers
mainstream users tend to only persue quick answers
i dont think this is a problem with the technology itself. you can always turn the temperature way up, use an exotic model, ask follow up questions, read (and interact) with the reasoning traces.
but how many mainstream users do this? why does it seem that LLM services dont encourage this?
i think that LLMs have a greater capacity for randomness and limitless generation of curiosity. and in some ways, raising temperature, and breaking down reasoning, is seen as an unnecessary flaw. you should never have to do this if the model isnt messing up.
the most ideal situation is… the model performs well. but the user asks crucial followup questions. i think if that isnt happening, then its the ecosystem(culture’s) fault.
leanleft@lemmy.ml 1 week ago
you could argue: two “problems” :
i dont think this is a problem with the technology itself. you can always turn the temperature way up, use an exotic model, ask follow up questions, read (and interact) with the reasoning traces.
but how many mainstream users do this? why does it seem that LLM services dont encourage this?
i think that LLMs have a greater capacity for randomness and limitless generation of curiosity. and in some ways, raising temperature, and breaking down reasoning, is seen as an unnecessary flaw. you should never have to do this if the model isnt messing up.
the most ideal situation is… the model performs well. but the user asks crucial followup questions. i think if that isnt happening, then its the ecosystem(culture’s) fault.