I discovered that my parents rely heavily on Doubao, asking it about everything and 100% trusting the answers it gives.
Doubao is really popular among middle‑aged and older people…
I tried to explain how large models work, but they said it was too complicated to understand. I’m looking for any popular‑science videos aimed at middle‑aged and older audiences so my parents will stop believing that everything a large model says must be correct.
This is not a simple issue for a large model; being able to trust the large model also means trusting Douyin and Baijiahao, and ultimately your goal will become helping your parents improve their cognition, which is extremely difficult.
You only understand after you’ve suffered a loss. Right now, people still believe because they haven’t experienced a loss yet; everyone is the same. Let’s see when we encounter the limits of large models—once we suffer a loss, we’ll realize.
There are many bugs where simple questions are answered incorrectly; you can look for them and demonstrate reproduction.
However, blindly trusting Doubao 100% indeed exceeds the decision‑making level brought by most people’s intelligence, representing an upgrade in decision‑making level.
Feels like what’s needed is 9.8 and 9.11 or output 150 x, that classic problem (
I don’t know if domestic AI has been specially tuned (
Indeed, at least it’s a bit stronger than Douyin and Baijiahao; even now I can still see college students using GPT to search for papers only to discover the results are fabricated 
Why not “using deep inquiry to write a National Natural Science Foundation grant” 
Oops, looks like I replied to the main thread.
But only when encountering situations like this can we realize the shortcomings of large language models. Teaching people through people is less effective than teaching them through things. Of course, if we could truly intervene in advance to prevent incidents, that would indeed be a good thing.