Don't you mean extinct?
This article draws a compelling parallel between CGI's revolution in film, which made stop-motion 'extinct', and LLMs' impact on software development careers. It addresses the growing anxiety among programmers about obsolescence, positioning AI as a tool for evolution rather than eradication. The author advocates embracing LLMs to enhance productivity and code quality, urging developers to adapt or risk being left behind in a rapidly changing landscape.
The Lowdown
The article "Don't you mean extinct?" opens with the iconic anecdote of Phil Tippett, the stop-motion master, being told his craft was 'extinct' after Industrial Light & Magic (ILM) showcased CGI's potential for Jurassic Park. This serves as a powerful metaphor for the current anxieties among programmers regarding the rise of Large Language Models (LLMs).
- The Extinction Analogy: Director Steven Spielberg initially favored stop-motion for dinosaurs, but a CGI proof-of-concept by Dennis Muren at ILM dramatically shifted his vision, leading Tippett to feel his career was obsolete.
- Evolving with LLMs: The author draws a direct line to today's software development, suggesting that rather than fearing obsolescence, programmers must 'evolve' by learning to use LLMs effectively, much like prior generations adapted to new technologies like the web or mobile.
- Learning Resources: The piece recommends resources like Andrej Karpathy's videos and Sebastian Raschka's book for understanding how LLMs work.
- Coding with LLMs: It argues that refusing LLMs will lead to falling behind, citing John Carmack's view that 'problem-solving is the core skill,' not just 'coding.' The author details how LLMs can assist in producing code while emphasizing the continued importance of human oversight for quality and architectural understanding.
- Enhanced Code Review: LLMs can significantly improve practices like writing commit messages, ensuring code clarity, breaking down large pull requests, and generating tests, thereby raising expectations for code quality.
- Smaller Teams & New Opportunities: The article posits that LLMs enable smaller teams to achieve more, fostering renewed interest in previously complex projects and opening avenues for exploring new technologies like
llama-cpporollama. - Motivation and Inspiration: Crucially, the author concludes by revisiting Phil Tippett's story, highlighting that despite his initial shock, Tippett adapted by co-developing the Dinosaur Input Device (DID), winning an Academy Award, and continuing a successful career. This underscores the message that adaptation and continuous learning are key to professional longevity.
The core message is one of adaptation and resilience. Rather than succumbing to fear, professionals should view technological shifts as opportunities for evolution, leveraging new tools to enhance their capabilities and maintain relevance.
The Gossip
Productivity Paradoxes
The article's strong assertion that developers refusing LLMs will 'fall behind' sparked considerable debate. Some commenters agreed, defining 'falling behind' as an inability to produce 'customer meaningful features' at a competitive velocity, suggesting LLMs act as 'pair programming with superman.' Others argued that for production-grade, high-quality code, LLMs don't necessarily increase speed; the time saved generating code is often reinvested in meticulous review, debugging hallucinations, and ensuring adherence to standards, creating new 'bottlenecks' in understanding generated output.
Openness vs. Oligopoly
A significant thread of discussion focused on the proprietary and paywalled nature of many leading LLMs. Some commenters voiced concern that relying on such tools goes against the traditional 'hacker' ethos of independent development and open access. While acknowledging the power of these models, they called for a fight for universal access and independent alternatives. Conversely, others argued that the immense capital required to train and run frontier models makes true independent development a pipe dream, creating an inherent 'resource play' that entrenches corporate control.
Quality Conundrums
The discussion extensively explored the trade-offs between quantity and quality when using LLMs. Critics expressed worry that AI often generates 'mediocre at best, on average it will be slop,' contributing to an unwanted abundance of low-quality digital content. They questioned if more output truly benefits society, which already yearns for 'more quality and less quantity.' However, a counter-argument emerged that LLMs can, in fact, facilitate higher quality by freeing up developer time for critical tasks like expanding test suites or focusing on architectural elegance, which might otherwise be sacrificed in the pursuit of speed.
Career Concessions
Commenters shared anxieties about job security and the evolving nature of software engineering, drawing parallels to past technological shifts in fields like VFX. Many acknowledged the inevitability of adapting to new tools, with some fearing their specialized skills might become 'utterly fucking pointless,' reminiscent of obsolete crafts. This theme also touched on the emotional aspect of such transitions, questioning whether increased productivity comes at the cost of personal joy or if the push for efficiency leads to unrealistic expectations and 'mental burnout' in the workplace.