The AI of Perchance’s character chat is stuck in 2024 and when asked what the current year is, it responds 2024. It also still thinks the US president is Joe Biden despite the current president being Donald Trump. I believe this topic has been brought up before but it makes no sense how it’s literally 2026 yet the AI is still stuck in 2024, literally two years behind. If you believe I’m complaining, I’m not. Also there’s zero excuse for this either. All that needs to be done really is to train the AI on modern up to date information (2025-2026). This doesn’t really have any significance otherwise, as it doesn’t affect roleplays. Just know that if you want an actual AI assistant, you’re better off using ChatGPT or Gemini.
You must log in or # to comment.
All AI models are restricted by their training data. Worth noting that this is also true of ChatGPT, Gemini, Claude, etc.
Two things make some experiences better than others:
-
A massive training/retraining pipeline to keep models as fresh as possible. This is insanely expensive in practice, so really only the big players do it.
-
RAG (retrieval augmented generation), which allows a model to interact with more current data using something akin to traditional search. These require significant additional infrastructure (and, they highlight some usefulness of LLMs as judges as opposed to knowledge bases).
The biggest players like ChatGPT and Gemini have both 1 and 2 in their pocket. Eventually the costs of constant retrainings won’t be covered by investor capital, so expect some differing results over time.
-
its called training data. the model was trained on data with a cut off point, in 2024 apparently. the ai doesnt learn new things, it only knows what its trained on
And its not like you can just add shit on top and boom, the AI continues to work. AFAIK making that bundle of information and having an AI work with it is a long and expensive process.
I believe there’s excuse and sense in those cases.
