Article pushes back on 'no one understands how LLMs work'
What happened
On October 12 an article on the Hacker News front page disputed claims in MIT Technology Review, TIME and WIRED that no one knows how AI works. The author argues LLMs are mechanical systems built from math functions humans designed and implemented, walking through tokenization, matrix multiplication and attention in the Transformer, and parameter updates via a loss function. Those steps can be inspected and reproduced, so the mechanism should not be described as beyond anyone's understanding.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageYes I Understand How LLMs Work
针对 MIT Technology Review、TIME、WIRED 等媒体“没人知道 AI 如何工作”的说法，作者指出 LLM 是由人类设计并实现的数学函数构成的机械系统，从 token 化、Transformer 中的矩阵乘法与注意力计算到损失函数调整参数，每一步都可被检查和复现。
Heat over time
Not enough continuous observations to draw a trend yet.