Subject: Factuality Bug: Wrong compressor wheel inducer diameter for Mitsubishi Lancer Evo 9 turbocharger (-01580)
Body:
Hello Gemini Team,
I found a persistent factual error in Gemini’s training data regarding the technical specifications of the Mitsubishi Lancer Evolution IX turbocharger (MHI part number: 49378-015.. / TD05HR-16G6C-10.5.).
Current/Incorrect AI Output: Gemini repeatedly states that the compressor wheel inducer diameter is 48.30 mm (which is an unverified aftermarket catalog value).
Correct Specification: The actual official compressor inducer diameter is 50.2 mm.
Official Reference/Source for Verification:This specification is officially certified in the FIA Homologation Form A-5688, specifically under Extension No. 05/03 VF for the Mitsubishi Lancer Evolution IX.
Please update the training dataset/knowledge base for this specific automotive component to reflect the official FIA motorsport homologation data rather than incorrect aftermarket catalog sheets.
Thank you!
Hello @Manfred_Marold ,
since static training data and knowledge cutoffs vary across model releases (see the Gemini Models Overview), could you please share which specific model version you tested?
Hello @Sai_Deepika_K ,
I’m in Germany and using the publicly available version 3.5 Flash. During chats, this version consistently states that the compressor inducer diameter is 48.3 mm instead of the correct 50.2 mm. It always apologizes for the incorrect information during the chat, but then forgets it again in the next chat. Another bug also occurs repeatedly: with two upgraded turbochargers from two different manufacturers… AGT 52mm “LE” Evo 4-9 turbocharger – Art Gennari Turbo LLC
and EVO IX 71HTA Super Core – Forced Performance
it also incorrectly claims in every chat that the compressor wheels have extended tapered tips. I would like to take this opportunity to request that this be corrected as well.
Thank you !
Hello @Sai_Deepika_K ,
will the relevant technical files be changed in the near future?
Thank you
Hello @Manfred_Marold ,
Depending on your use case, here are the recommended approaches to address this:
- If you are the manufacturer or data owner: Ensure that up-to-date, official documentation is publicly available and that outdated or conflicting technical specifications are removed. This ensures future web crawls and index updates reflect the correct information.
- If you are building an application that requires strict factual precision: We strongly recommend leveraging Retrieval-Augmented Generation (RAG) so that the model relies directly on your verified reference documents rather than general training data. You can implement RAG using the Gemini API File Search tool, which allows you to upload official manuals and restrict responses strictly to your source files.
We are continuously working to improve base model accuracy, and we appreciate your feedback!