Ai recommendations on engine improvements

[JoeMag]

Has anyone tried this? I continue to be amazed at what these tools can do.

I decided to feed ChatGPT my engine specs (bore, stroke, cam data, exh data, intake data, head flow, engine type, dyno, etc). Asked it what could be done to improve particular parts of dyno or generally raise Hp. It gave me some cam recommendations that were in the direction of a new cam Elgin just designed for me. Also, had header recommendations that were in the direction of PipeMax calculations.

A valley in dyno did not align with what it expected based on engine data and it provided possible cam, intake, and exh interactions that may have caused it.

I have some new headers to try... Plopped the data for those in and had it estimate dyno plot with these new headers.

Basically fed it pictures of all the data except for the very simple ones.

[silverlock]

Yes, but AI models WILL “hallucinate” seemingly valid but incorrect data — which can be hard to know.

I find ensuring web search is on + checking sources to validate any findings is a near-must. I was rubber-ducking an axle flange seal leak on my SC+915/61. It kept suggesting it was a 50mm seal (flat out wrong), linked to a source that said nothing of the sort, and claimed the PET was updated later (also untrue).

I only knew it was wrong because I had done research elsewhere. Not a bad model either (this was OpenAI’s o3 model!). Not the only time it’s done that but it was the most egregious.

[JoeMag]

Totally agree... it's definitely better to be an "informed observer" and check if you can.

I had it mix up OD and ID, but when I clarified, it fixed it.

I recently asked it to do some compression ratio calcs based on some parameters changing and was interesting to watch it self iterate. It came up with an answer, said that can not be right, and went back and iterated a couple times before giving final answer.