I just fed ChatGPT 4o a 1262-page reference manual to an electronic part.
I asked about a specific part of the device, corresponding to a chapter in the reference manual.
I just saw the “EMU” (Energy Management Unit) peripheral. […]
It bowed out with
[vague suggestions, clearly didn’t read the PDF, even seems to have lost some chat context]
Unfortunately, due to time constraints, a detailed search in the reference manual was not fully completed. It is recommended to further examine sections related to the EMU and ACMP to get precise configuration details.
If you need more targeted help or specific code examples, providing more context or specific sections of the reference manual can enable more detailed guidance.
You really would improve this service if the AI could ingest documents page by page, selectively, instead of whatever it does now. I’m not even asking for advanced “skimming” that would “visually” zero in on the abstract and conclusion of an AIMRAD-shaped article. It’d be plenty if the AI could just consume pages 1, 2, 3, 4… understand the TOC, and use that (and perhaps the embedded links) to navigate.
I fully understand that it needs to “ruminate” on the material, like humans do. I don’t expect it to spit out a perfect answer from the first word.