{
  "id": 20558002,
  "title": "Pretraining vs post-training: one apple problem, four learning signals",
  "url": "https://urgent.news/2026/11/18/pretraining-vs-post-training-one-apple-problem-four-learning-signals",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-11-18T06:50:22.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/drvexo_llm/pretraining-vs-post-training-one-apple-problem-four-learning-signals-5h46"
  },
  "original_language": "en",
  "account": "An LLM's knowledge originates from pretraining, where it learns word meanings, associations, patterns and basic problem-solving skills through exposure to large text and code datasets. The model learns to predict subsequent text, allowing knowledge and abilities to emerge from the parameter structure rather than a literal document store.\n\nPost-training refines these skills by teaching the model to follow instructions correctly. For example, if asked \"Show the equation before giving the answer,\" the model can learn to produce the correct response (2 + 3 = 5) while also explaining how it arrived at that solution.\n\nTwo training methods exist to shape the model's instruction-following abilities: InstructGPT uses demonstrations of good responses, while DeepSeek-R1 trains from preference pairs. Both methods aim to optimize relative preference rather than guarantee absolute probability improvements.\n\nTo ensure the model's answers are accurate, reward-based reinforcement learning techniques like verifiable rewards or classic RLHF are employed. These methods involve sampling during training, checking answers or executable tests, and providing feedback to the model.\n\nPretraining and post-training are not mutually exclusive and can both enhance an LLM's capabilities. However, neither method guarantees correct answers for unseen problems. The accompanying animation, with Chinese captions, serves as an illustrative teaching tool rather than an exact representation of actual LLM training results.",
  "summary": "Where does an LLM’s knowledge come from—and what changes after pretraining? Follow one example: a child has two apples and buys three more. Then add a request: “Show the equation before giving the answer.” Pretraining builds foundations Training on large collections of text and code can teach word meanings, factual associations, language patterns and some problem-solving abilities. For…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}