← All Demos

Base Model vs Instruction-Tuned Model

Same architecture, different training — see why post-training matters.

robert@barcik.training

Key insight: The base model is an internet text simulator — it autocompletes based on patterns in training data. The instruction-tuned model (SFT → RLHF) learns to follow instructions and be helpful. Same underlying knowledge, fundamentally different behavior. InstructGPT (Jan 2022) demonstrated this; ChatGPT (Nov 2022) brought it to the public.
USER →
Base / Foundation Model pre-training only
Click "Generate" to see output...
Instruction-Tuned Model SFT + RLHF
Click "Generate" to see output...
AI transparency

This demo is a scripted, self-contained browser simulation. Nothing you type is sent to an AI model and no live AI system runs behind it, even where it plays one. Its code and copy were built with generative AI (Anthropic’s Claude) and reviewed by Robert Barcik, who is responsible for what is published (LearningDoe s.r.o.). Disclosed in the spirit of Article 50 of the EU AI Act.