A type of phrasing favored by AI, 'contrast framing,' uses an 'it's not this, it's that' structure

Overview Debates Related claims Topics

Nathaniel Whittemore prompts artificial intelligence models to steel-man opposing viewpoints when evaluating complex topics

Prompts for GPT Tutor in a Turkish high school math study exceeded 500 words, compared to approximately 50 words for standard GPT-4

Prompts are a key unit of post-training for AI models

Early AI models performed better on tasks after being given a 'pep talk', a behavior that would have been considered an insane prediction 20 years prior

People are increasingly using AI-like phrasing, or 'prompt speak,' in professional and casual conversations

'Caveman speak' is characterized by very short, direct phrases

Google searches for the term 'token maxing' peaked in June

Using AI-influenced speech patterns can make a person's communication sound generic

People are mimicking the direct, cost-effective phrasing used in AI prompts, known as 'caveman speak.'

An essential skill for data practitioners is testing how different data contexts affect an AI model's output for the same prompt

Practitioners should test AI models by sending different data contexts for the same prompt

The SCQA (Situation, Complication, Question, Answer) model from a Heavybit executive communications workshop is a recommended framework for advanced prompting

Effective use of Claude Code requires building a mental model of the agent's operational patterns

The "Explain it like I'm five" skill uses the key phrase "big picture is few words" to generate concise summaries

A key skill in agentic coding is identifying and articulating unknown variables and requirements for the agent

Learning the specific vocabulary of a domain improves the effectiveness of prompting Claude

Prompting is a critical meta-skill for working effectively with Claude Code

A prompting technique to improve agent performance is to ask the agent to create "decision notes" or "implementation notes."

Custom instructions in files like Claude.md need to be tailored to specific model versions, such as Fable 5.0 versus Fable 5.1

Including a long history of failure modes from older models in a Claude.md file can over-constrain newer models