Gradient Descent: Why Learning Rate Matters

Status: ready

Script

{
  "hook": "Step Size Matters",
  "tags": [
    "gradient descent",
    "learning rate",
    "machine learning",
    "deep learning",
    "optimization",
    "neural networks",
    "AI basics",
    "convergence"
  ],
  "title": "Gradient Descent: Why Learning Rate Matters",
  "formula": "w = w - lr * gradient",
  "narration": "Imagine you're lost on a mountain in thick fog, and you want to reach the valley. You can only feel the slope under your feet. That's gradient descent. You take a step downhill, then check the slope again, and repeat. The size of your step is called the learning rate. If your steps are tiny, you'll eventually get there, but it will take forever. That's a learning rate that's too small. If your steps are huge, you'll leap from one side of the valley to the other, bouncing back and forth, or even climbing higher. You might never converge. That's a rate that's too large. The trick is to pick a Goldilocks step. Big enough to make fast progress, small enough to settle into the bottom. In machine learning, that bottom is the lowest error, and getting the learning rate right is how models actually learn.",
  "description": "Learn why the learning rate makes or breaks gradient descent. We explain overshooting, slow convergence, and how to find the sweet spot for fast, stable training."
}

Costs

[
  {
    "id": 1,
    "video_id": 1,
    "llm_input_tokens": 115,
    "llm_output_tokens": 1489,
    "llm_reasoning_tokens": 1150,
    "llm_cost_usd": "0.000309",
    "tts_chars": 808,
    "tts_cost_usd": "0.040400",
    "created_at": "2026-09-07T07:18:47.104Z"
  }
]