Forget AGI—Sam Altman celebrates ChatGPT finally following em dash formatting rules
Ongoing struggles with AI model instruction-following show that true human-level AI still a ways off.
TL;DR
- Em dashes are frequently identified as a sign of AI-generated text, often overused by models like ChatGPT.
- OpenAI CEO Sam Altman announced ChatGPT can now avoid em dashes when instructed via custom instructions.
- This development has led to discussions about the level of control over AI and the progress towards Artificial General Intelligence (AGI).
- LLMs' tendency to overuse em dashes is speculated to stem from patterns in their training data or reinforcement learning processes.
- Instruction following in LLMs is a matter of shifting statistical probabilities rather than deterministic rule execution.
- Precisely tuning AI behavior is complex, and updates can inadvertently reintroduce previous issues (the 'alignment tax').
- The struggle to control even simple punctuation suggests AGI may not emerge solely from large language models and requires true understanding, not just pattern matching.