@monteluna

Karpathy released this interesting microgpt code that gives a look into how LLMs work under the hood. The interesting part: "What’s the deal with “hallucinations”? The model generates tokens by sampling from a probability distribution. It has no concept of truth, it only knows what sequences are statistically plausible given the training data." https://karpathy.github.io/2026/02/12/microgpt/

0 reply

0 recast

6 reactions