Practical Observations on LLM Capabilities and Limitations After Two Months of Daily Use

Tomasz Tunguz shares insights from two months of daily LLM usage across data analysis, coding, and web search. He identifies that incremental coding guidance outperforms comprehensive task descriptions, LLMs struggle with debugging their own errors, and LLM-based search could replace Google for summarization but not exhaustive queries, though recency of training data remains a critical limitation.