Comment on GPT-4-turbo preliminary benchmark results on code-editingparentComments−crooked-v2yFrom the anecdotes there, it sounds a whole lot like something is going on where it's internally doing a summarizing step to fit into a much smaller actual context window than 128K.
Comments
From the anecdotes there, it sounds a whole lot like something is going on where it's internally doing a summarizing step to fit into a much smaller actual context window than 128K.