Scaling Laws and Chinchilla

Hard

llms DeepMind Google OpenAI

Multiple Choice

Larger models always outperform smaller models regardless of training data Model size and training tokens should be scaled proportionally for a given compute budget, and many LLMs were undertrained Only training data size matters; model size has diminishing returns beyond 1 billion parameters Compute-optimal training requires training for exactly one epoch over the data

Correct!

Incorrect — try again next time!

Upgrade to Pro to unlock solutions and explanations