some high level reflections now that I’ve sufficiently reviewed the output of Deep Research and compared it against my own desk research:
the output is too long, does too much “interpreting”, and is really weird (though not wrong exactly) about sources.
the output is too long, does too much “interpreting”, and is really weird (though not wrong exactly) about sources.
Reposted from
justin from the internet
having used several of them now I just want to make sure my fellow AI skeptics/haters are aware that there are _drastic_ differences between various the models and expressions/applications of those models
Comments
I cannot stress enough how many times this long ass document linked to the 2024 stack overflow developer survey lol
in one section, instead of citing the github developer survey directly (a source referenced frequently in the output), the output summarized the source correctly but cited a third party content blog that cited the original survey
I sifted through 40 pages of what was essentially the output of a chaos goblin intern