Ai Chatbot Evolution Comparison

The question

Two generations of conversational AI, sixty years apart: a rule-based ELIZA (1966 design) and a modern instruction-tuned language model. What does each actually do better, side by side, on the same input?

How it works

• ELIZA (eliza.py) — twelve regular-expression rules, each with response templates, and a reflection engine that turns "I" into "you" and "my" into "your".
• The language model (LLM.py) — Qwen2.5-1.5B-Instruct run locally through a Hugging Face Transformers pipeline.
• The comparison (chat_comparison.py) — a Tkinter desktop app with the two conversations side by side. One message goes to both; ELIZA answers immediately while the model generates on a background thread, so the window never freezes.

What it shows

• ELIZA is instant and fully predictable, but it only recognises the patterns it was given.
• The model handles open-ended input, at the cost of a large download, slower replies and answers that can't be predicted in advance.
• The trade-off — predictability against flexibility — is the same one a team faces today when choosing between a rules engine and an LLM for a narrow task.