Linus Torvalds, the creator of Linux, has provided a candid reassessment of artificial intelligence's impact on programming productivity. Speaking at the Open Source Summit India 2026, Torvalds walked back his earlier claim that large language models could deliver a tenfold increase in developer efficiency, admitting that figure was 'pulled out of my arse' and lacked scientific basis.
The Linux kernel maintainer revealed that the open-source community has been grappling with a flood of AI-generated code submissions of questionable quality. Until early this year, the volume of machine-generated rubbish significantly outweighed genuinely useful contributions. Torvalds explained that poorly crafted AI patches consume considerable human resources, as experienced developers must invest substantial effort verifying whether automated bug reports are legitimate or simply algorithmic hallucinations.
Torvalds emphasised that effective use of AI tools requires human oversight and engagement. He stressed that developers cannot simply use LLMs to identify bugs, generate reports, and submit them without further involvement. The kernel team expects contributors to provide suggested patches and engage in iterative refinement. Many AI-generated fixes, he noted, function as superficial 'band-aid' solutions that address immediate symptoms whilst leaving underlying vulnerabilities lurking, ready to resurface elsewhere in the codebase.
Whilst acknowledging AI's limitations for kernel-level development, Torvalds does find value in using LLMs for personal prototyping projects. He employs them to rapidly test concepts, though the generated code typically requires substantial revision before becoming production-ready. Despite the frustrations, he welcomes AI-discovered security vulnerabilities, even when they prove embarrassing by revealing decades-old bugs. His philosophy rejects shooting the messenger in favour of transparency and improvement.
On a lighter note, Torvalds shared his creative use of AI for personal entertainment, particularly using generative tools to insert Godzilla into travel photographs he sends his children. He concluded his remarks by observing that artificial intelligence offers both valuable and frivolous applications, suggesting that 'Godzilla is a great place to stop' the conversation.
Fuente Original: https://linux.slashdot.org/story/26/07/12/2053201/linus-torvalds-on-ai-junk-patches-humans-and-godzilla?utm_source=rss1.0mainlinkanon&utm_medium=feed
Artículos relacionados de LaRebelión:
- AI Coding Agents Vulnerable to Bash Exploits
- Coheres North Mini Code Open-Source Agent For Coding
- Microsoft Unveils RTX Spark Desktop and Linux Tools
- Linus Torvalds Reveals AIs Impact on Linux
- RHEL 102 Launches AI-Powered Command Line Tools
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario