J’ai arrêté d’écrire mes prompts

J’ai arrêté d’écrire mes prompts

I stopped writing my prompts

Title history
  • J’ai arrêté d’écrire mes prompts — observed on 20261101-0050 UTC (YouTube)

by DevArt · View the channel on PSI

Humanities, Social Sciences & Thought AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 DevArt 👥 19K 📅 August 20, 2026 ⏱ 13 min 👁 6K 📄 expert opinion 🧭 2026-11-01
Available in: English (current) Français

Keywords

voice inputLLMprompt engineeringworkflowvoice coding

Summary

The creator shares his recent shift from typing prompts to using voice input when interacting with AI models. He explains that after seeing a tweet by Andrej Karpathy advocating voiced monologues to provide more context, he tested the method and found it significantly more efficient. He claims that speaking allows him to express 2-3 times more words per minute than typing, and that modern LLMs are trained to understand spoken reasoning, including hesitations and emotional emphasis. He recommends a hybrid approach: using voice to generate a draft context, then editing it before sending. He also reviews several voice-to-text tools, particularly highlighting VoiceInk as an open-source, privacy-oriented, one-time-purchase option. He mentions that OpenAI’s model guidance suggests giving models more context, which voice naturally provides. He concludes by asking viewers for their opinion and encourages adopting voice-driven workflows for coding and brainstorming.

140 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers practical value for AI users by presenting a novel workflow modification backed by anecdotal evidence and a reference to Karpathy’s tweet. The argument is built on personal experience and a cited external guidance, but it lacks systematic testing or broader empirical data. The reasoning is coherent and addresses common pain points, but the lack of critical examination of potential downsides (e.g., context overloading, accuracy issues) weakens the argumentative depth.

Scientific Rigor, Source Quality, Title Accuracy

The main sources cited are the OpenAI API guidance and OpenAI’s GPT-4o voice introduction, both relevant. The creator also mentions several third-party tools (WhisperFlow, Super Whisper, VoiceInk) but does not provide independent evaluations. The title accurately represents the content, with no clickbait. The video includes a promotional segment for VoiceInk, though it’s not sponsored; the creator explicitly denies sponsorship. Overall, the scientific rigor is moderate, relying on personal testimony and corporate documentation rather than peer-reviewed studies.

163 words

Title / Content Match

The title accurately reflects the content, which discusses abandoning traditional text prompts in favor of voice interaction with AI.

Quality & Reliability

6/10

The video presents a personal experience-based workflow shift, supported by references to OpenAI guidance and a tweet from Andrej Karpathy, but lacks formal scientific evidence. The advice is practical and coherent, though subjective and product-flavored.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

  • OpenAI GPT-4o blog — Supports the claim that modern models handle voice interaction effectively.

External References

Contribution & Novelties

This video contributes by popularizing the concept of voice-first prompting, a trend observed in the AI community, and provides a practical workflow comparison. It highlights the increased context richness of spoken language and suggests a hybrid editing approach.

Pour aller plus loin :

  • Voice User Interfaces — Overview of voice interaction technology.
  • Speech-to-Text API — Background on speech recognition technology.
  • Andrej Karpathy’s tweet — The original tweet referenced (exact URL not available, but tweet is known).

76 words

Radar Profile

The radar profile shows moderate scores across all dimensions, with a slight strength in information relevance and a weakness in technical depth, reflecting a practical but not deeply rigorous presentation.

Reliability 6/10