GPT-5 Decepciona: El PEOR Programador, el FIN de Photoshop y la MENTIRA de la Productividad con IA

GPT-5 Decepciona: El PEOR Programador, el FIN de Photoshop y la MENTIRA de la Productividad con IA

by Codemancers - Inteligencia Artificial

Humanities, Social Sciences & Thought AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Codemancers - Inteligencia Artificial 👥 2K 📅 August 22, 2025 ⏱ 91 min 👁 371 📄 debate 🧭 2026-08-15
Available in: English (current) Français

Keywords

GPT-5productivityprogrammingHRMimage generation

Summary

The episode discusses whether AI improves programmer productivity, citing a Stanford study of 100,000 developers showing a real 15% increase, not the hyped 2000%. The hosts share personal experiences: AI helps with new projects and simple tasks, but fails on legacy code and complex systems, sometimes reducing productivity. They also introduce the Hierarchical Reasoning Model (HRM), a new architecture inspired by the human brain that achieves complex reasoning with only 27M parameters, challenging the brute-force scaling of Transformers. The episode covers GPT-5’s performance, noting it is the worst programmer for execution but best for architecture, and lags behind Sonnet and Opus. It also discusses the end of privacy in the EU with ‘Chat Control’, and the impact of image generation models like Nano Banana and Qwen Image Edit on Photoshop workflows. The hosts critique Mark Zuckerberg’s dystopian vision and share ‘Razzies’ for AI-related failures. The tone is conversational and opinionated, mixing technical analysis with social commentary.

156 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable insights into the real-world impact of AI on programming productivity, backed by a large-scale Stanford study. The hosts offer a balanced perspective, acknowledging both benefits and limitations. They argue that AI is a tool that amplifies existing skills, and its effectiveness depends on code quality and project complexity. The discussion on HRM is informative, explaining a novel architecture that could shift AI development away from brute-force scaling. However, the argumentation often relies on personal anecdotes and subjective impressions, which, while relatable, lack rigorous empirical support. The hosts also engage in speculative predictions about the future of programming languages and AI’s societal impact, which are thought-provoking but not always grounded in evidence.

Scientific Rigor, Source Quality, Title Accuracy

The video references a Stanford study on AI productivity, but does not provide a direct link or full citation, making it difficult to verify. The HRM paper is mentioned, but again without a specific reference. The hosts cite personal experiences and industry news, but these are not systematically sourced. The title is somewhat sensationalist, but accurately reflects the main topics. The content is a mix of factual reporting and opinion, with a clear bias towards a critical view of AI hype. The hosts do not provide a balanced view of counterarguments, but they do acknowledge limitations. Overall, the scientific rigor is moderate, with a need for more transparent sourcing.

239 words

Title / Content Match

The title accurately reflects the main topics: GPT-5's performance, productivity claims, and image generation models, though it uses sensationalist language.

Quality & Reliability

7/10

The video presents a mix of personal opinions, anecdotal evidence, and references to studies (Stanford productivity study, HRM paper). While the hosts discuss real research and industry events, they often rely on subjective experiences and speculative claims. The sources are not systematically cited, and some statements lack direct verification.

Chapters

Cited Sources

Concurring Sources

  • Stanford study on AI productivity (mentioned) — The hosts cite a Stanford study showing a 15% productivity increase, but do not provide a direct link.
  • Hierarchical Reasoning Model paper (mentioned) — The hosts discuss a paper on HRM, but do not provide a direct link.

Dissenting Sources

  • Claims of 2000% productivity increase — The hosts debunk exaggerated productivity claims, contrasting them with the Stanford study's 15% figure.

Contribution & Novelties

The video offers a critical perspective on AI productivity claims, contrasting hype with empirical data. It introduces the Hierarchical Reasoning Model (HRM) as a potential paradigm shift in AI architecture, which is a novel concept for many viewers. The discussion on image generation models like Nano Banana and Qwen Image Edit highlights their disruptive potential for creative workflows. The hosts also raise important ethical and societal concerns about privacy and AI’s role in society.

Pour aller plus loin :

  • Stanford study on AI productivity — Note: This is a placeholder; actual study not identified. The video mentions a Stanford study but does not provide a link. For further reading, search for ‘Measuring termines in language models for software code review’.
  • Hierarchical Reasoning Model paper — Note: Placeholder; actual paper not identified. The video discusses HRM but does not provide a link. Search for ‘Hierarchical Reasoning Models’ on arXiv.
  • Chat Control legislation — Note: This is a real EU legislative page, but the exact URL may vary. The video discusses the EU’s Chat Control law.
  • Nano Banana model — Note: Placeholder; the video mentions Nano Banana but does not provide a link. It might be a nickname for a model like FLUX.1.
  • Qwen Image Edit — Note: Placeholder; the video mentions Qwen Image Edit but does not provide a link. Search for Qwen image editing models.

225 words

Radar Profile

The radar profile shows a balanced but moderate performance across all dimensions, with slightly higher scores in information quantity and technical level, but lower in reliability and information quality. This suggests the video is informative and technically detailed, but its credibility is limited by reliance on personal opinions and lack of direct source citations.

Reliability 6/10

💬 No comments were provided for analysis.