Search Results
6/30/2025, 5:31:32 PM
>>509130963
>I can't find any information about nthe case you're referring to
Because I made it up, but that won't stop the LLM from doubling down that not only does the case that I made up exist, and even telling me with absolute confidence that my second-guessing about my recollection of the case I made up is mistaken.
LLMs don't think or reason, they are just a much, much more complicated version of fitting a line to data; and if you tell a computer to fit a line to data and use that line to interpolate new results - those results will inevitably be on that line, whether or not the data you fed the model is actually linear. It's this same kind of logic that explains why LLM responses skew towards the sycophantic. It's a predictive model that's going to roll with whatever inputs you feed it and will always tend to trend along that initial trajectory.
>I can't find any information about nthe case you're referring to
Because I made it up, but that won't stop the LLM from doubling down that not only does the case that I made up exist, and even telling me with absolute confidence that my second-guessing about my recollection of the case I made up is mistaken.
LLMs don't think or reason, they are just a much, much more complicated version of fitting a line to data; and if you tell a computer to fit a line to data and use that line to interpolate new results - those results will inevitably be on that line, whether or not the data you fed the model is actually linear. It's this same kind of logic that explains why LLM responses skew towards the sycophantic. It's a predictive model that's going to roll with whatever inputs you feed it and will always tend to trend along that initial trajectory.
Page 1