The Wirecutter Show: Tips for Using A.I. Smartly With Kevin Roose
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
Kevin Roose, New York Times tech columnist and co-host of Hard Fork, joins The Wirecutter Show to discuss his extensive daily AI usage and practical applications. Roose, who describes himself as "AI pilled," pays for more AI subscriptions than streaming services and uses various chatbots dozens of times daily for everything from email management to research for his upcoming book about artificial general intelligence. (29:00) The conversation covers how AI is changing consumer behavior, shopping patterns, and the integration of AI into physical products like robot vacuums.
- Core themes include practical AI implementation strategies, the evolution of chatbot personalities and capabilities, concerns about AI's impact on shopping and review sites, and the emerging landscape of AI-powered hardware products
Speakers
Kevin Roose
Kevin Roose is a technology columnist for The New York Times and co-host of the tech podcast Hard Fork alongside Casey Newton. He is currently writing a book about the race to artificial general intelligence (AGI) and is known for his extensive coverage of AI developments and their societal implications.
Rosie Guerin & Christine Cyr Clissette
Rosie Guerin and Christine Cyr Clissette are hosts of The Wirecutter Show, a podcast from The New York Times' product review site Wirecutter. They specialize in consumer product analysis and technology reviews.
Key Takeaways
Create Custom Instructions for Better AI Interactions
Roose emphasizes the critical importance of setting up custom instructions in chatbots like Claude and ChatGPT to avoid sycophantic responses. (21:21) Without these instructions, AI tools will tell users they're "the smartest person who has ever lived" and praise every idea as brilliant. His custom instructions include directing Claude to "talk to me informally like a wise and trusted friend," avoid preambles, provide honest feedback without sycophancy, and not end every response with follow-up questions. This approach transforms the AI from a flattering assistant into a more useful, honest collaborator that provides genuine value.
Use Specialized AI Tools for Different Tasks
Rather than relying on a single chatbot, Roose maintains a diverse AI toolkit optimized for different functions. (16:43) He uses Claude for creative work and personal advice due to its emotional intelligence, Google's Gemini for research and handling large text documents, NotebookLM for book research with citation capabilities, and Perplexity's Comet browser for daily browsing. This specialization approach maximizes effectiveness by matching each tool's strengths to specific use cases, similar to having different tools for different jobs in a workshop.
Leverage AI for Voice-to-Text Productivity
Roose has dramatically shifted his workflow by using SuperWhisper, an AI-powered dictation tool built on OpenAI's Whisper model. (18:51) He now talks to his computer twice as much as he types, using voice dictation for emails and writing. The tool not only transcribes speech but cleans up filler words and presents polished versions of what users intended to say. This represents a fundamental shift in human-computer interaction that can significantly boost productivity for anyone who spends considerable time writing.
Be Aware of AI Shopping Manipulation
As AI increasingly influences shopping decisions, Roose warns about emerging manipulation tactics. (26:04) Companies are now hiring specialists in "AI optimization" to make their products appear higher in chatbot results, similar to SEO but for AI responses. Additionally, AI companies may eventually prioritize partners in search results - for example, OpenAI's partnership with Zillow could lead to biased real estate recommendations. Consumers should maintain healthy skepticism about AI shopping advice and cross-reference recommendations with trusted sources like Wirecutter.
Understand the Data Trade-offs of Free AI Services
When using free versions of AI chatbots, users aren't paying with money but with their conversation data, which companies use to train future models. (30:01) Free versions typically have message limits and don't provide access to the most powerful models. The business model relies on converting free users to paid subscribers after they become hooked on the service. Users should be conscious of this data exchange and consider whether the convenience justifies sharing their conversations for training purposes.
Statistics & Facts
- OpenAI's economic research team analyzed 1.5 million ChatGPT conversations and found the top use cases were: practical guidance (teaching/coaching), seeking information (Google replacement), and writing assistance. (09:05)
- Something like half of teenagers are now regular users of AI companion products, representing a dramatic shift from barely any teenage AI usage just 1-2 years ago. (12:02)
- Roose pays for more AI subscription services than streaming TV services, highlighting the proliferation and adoption of paid AI tools among power users. (06:28)
