Well this really p*ssed me off
Summary
AI summaries can be incomplete or wrong. Verify anything important against the original video.
Theo rants about the difficulties of using LLMs for code generation, particularly when dealing with complex codebases and the limitations of current models in providing accurate and reliable output. He highlights issues with code quality, cost, and the need for human oversight.
The video features Theo expressing his frustration with LLMs in software development, especially concerning code quality and cost. He begins by discussing the challenges of using LLMs for code generation, noting that current models struggle with complex codebases and can produce unreliable output, leading to issues like memory leaks and crashes. He highlights the high cost of using advanced models like Claude Fable 5 and the difficulty in obtaining reliable results.
Theo then delves into the problems Jarred, a Zig community member, faces with his open-source project, Bun. He criticizes Jarred's approach and management style, suggesting it's a "beginner energy" approach that leads to mediocrity and poor communication. He recounts a specific incident where Jarred's actions were perceived as condescending and unrealistic, and highlights that Zig community members were eager to find work coding in Zig on the clock, but Jarred's approach caused issues.
The video also touches upon the challenges of LLM-generated code, mentioning the existence of "unsafe" blocks and the need for developers to understand these limitations. Theo questions the reliability of LLMs for production environments and the costs associated with their use, citing the example of a $165,000 API pricing for certain usage.
Finally, Theo shares his personal experience with a Rust rewrite of Bun, noting its success and performance improvements, but also acknowledging the ongoing challenges and the need for continuous refinement. He concludes by emphasizing that while LLMs are powerful tools, human oversight and critical thinking remain essential in software development.
Claims & arguments
LockedKey Points
LockedWorth watching if: This video is for developers interested in the challenges and potential of using LLMs for code generation, as well as those curious about the development process of projects like Bun. It's particularly relevant for anyone working with large codebases or considering AI-assisted programming.
Sign in to unlock the full extract
Every claim, key point, and timestamp for this Theo - t3․gg video — plus a daily email of every channel you follow.
Sign in with GoogleNo credit card. Free tier forever.