In June 2023, a New York lawyer filed a brief citing six court cases. None of them existed — an AI had invented them all. This episode opens the black box to show why large language models make things up with total confidence, and how to catch it.
What's inside:
- What a language model actually does when you ask it a question (hint: it doesn't store facts — it predicts the next word)
- Why the same machinery produces true answers and fabricated ones, and why the model can't tell them apart
- Why hallucinations compound and why regenerating gives a different answer each time
- Three practical rules to catch hallucinations: verify anything specific, distrust suspiciously perfect answers, ask twice in a new chat
Blackbox opens one piece of the AI machine every week — five-minute animated explainers, no hype, no jargon left unexplained.
Next episode: What Happens When You Type a Prompt?
Visuals made with AI tools and fact-checked by a human. Voice-over: AI-generated.
Blackbox (@OpenBlackBox)
ข้อเสนอแนะ
รายงาน
ห้ามทำซ้ำหรือดัดแปลงโดยไม่ได้รับอนุญาตจากครีเอเตอร์
ข้อเสนอแนะ
รายงาน