How AI learned to chat
4:35 · Horizontal
Sounding right isn't being right
AI sounding sure of itself doesn't make it right. We start with how it learned to talk.
It opens with AI announcing, out of nowhere, that there's a 15-day holiday coming. Xiaoming nearly believes it and drags Manman off to pack. Xiaoting stops them: don't trust it just yet.
Next comes a cat-spotting experiment starring Manman. With only three rules (pointy ears, whiskers, a tail), it can't recognize Manman the moment she hides her tail. Show the machine lots of labeled examples instead, and it slowly picks up the pattern on its own. That's machine learning. The episode also covers tokens, and why what you said earlier shapes how it carries on.
The slip-up comes at the end: however nicely it's put, with nothing to back it up, "15 days" has no basis. Confidently made-up answers like this are often called hallucinations. One thing to take away: check where important numbers come from, and run big multiplications through a calculator.
It confidently made up numbers, and they sounded pretty convincing.
Say thisSwap the [ ] part for yours
Look up [what you want to know] for me. After each fact, add the source link and its date. If you can't find something, just say so. Don't guess.
