Rust Author Points Out Limits of LLM Learning, Argues for Teacher Lineage

An essay on a personal blog by the author of a Rust technical book is attracting attention on a developer forum. The theme is the difference between large language models and human instruction in learning support. Coverage of bastiangruber.ca by gruberb on Lobsters describes the author’s own learning experience and his experience revising his book. It contrasts the appeal of efficient explanation generation with the problem that the provenance of answers is unclear.

The author reveals that he is a 37-year-old remote worker. He says he never fit into the framework of school, but has always loved learning itself. He says he was strongly drawn to the emergence of language models that can produce diverse explanations from just a few sentences of input. He argues that even answers based on statistical prediction can advance understanding when combined with proper reasoning and repeated dialogue. Imagining using them at home to dig into topics he could not ask about in class, he notes they could have helped his studies. On the condition that the regulatory environment is put in place, he says he leaned toward seeing language models as a pure benefit.

Embracing LLMs for Learning Support

The author initially valued language models highly as companions for learning. He saw them as effective for shy students who cannot ask follow-up questions and for catching up on lessons where they had lost focus. In an environment prone to isolation through remote work, he says their value as partners for exchanging ideas is great. He states that even when they contain errors, the dialogue itself creates insights. He expressed the view that, if systems for data-centric construction, training, and use are established, they would be beneficial on the whole. He feels that in development work as well, the speed of self-study and experimentation has increased.

This positive assessment changed after encountering a book on meditation and Buddhism. The book in question is Susan Piver’s “Inexplicable Magic.” The book is not about language models or development methods. Even so, a passage about the provenance of learning shook the author’s position. He says he felt that the question of who you learn from also applies to acquiring technical skills. He is turning his attention to elements that cannot be measured by speed and volume of explanation alone.

Doubts About Generated Answers Without Lineage

A quotation from the book is placed at the core of the essay. The point is that we should be aware of the doorway into learning and the existence of the people who have passed through it. The passage quoted from the original is as follows.

Who you learn from is important because you have entered the practice via a certain doorway through which many others have passed.

A further passage is introduced that questions whether there is a lineage behind the instructor and their instructions.

Is there a lineage behind your instructor and their instructions? If so, good, because we don’t want any made up bullshit.

The author reinterprets language model responses against this standard. He describes them as plausibly made-up content, with no person standing behind them and no traceable provenance. He argues that persuasive wording and well-grounded instruction are different things. He concludes that we should learn from people who reject nonsense.

This point concerns the verifiability of generated output. If no record remains of who said what, when, and on what basis, no path for correcting errors can be built. The approach of tracing the lineage of a public implementation also overlaps with the整理 of the driver-layer revival covered in Vulkan Video Encoding Revived on Intel Alchemist GPUs. In development where change history and responsible persons are clear, isolating problems moves forward. He argues that the challenge is that single-shot answers from language models come with no such history attached.

How Teachers and Apprenticeship Shaped His

Professional View

The author reflects on his own professional formation. He says teachers, senior engineers, and managers who influenced him loomed large. He states that guidance from those around him supported his growth as much as his own efforts. He says he inherited not only his view of the industry and his profession, but also his way of being as a person. He expresses sympathy for the book’s suggestion to hold in mind a person during meditation. He takes it as an act of affirming gratitude and connection.

He also touches on his experience of apprenticeship as an electrician. He says there was a relationship in which he learned by watching senior colleagues’ techniques on site and directly asking them why failures happened. He argues that undocumented judgment criteria and attitudes toward safety are conveyed through face-to-face instruction. The move to clarify the provenance of testing infrastructure also resonates with the testing culture shown in GNOME OS Test Center, Inspired by Apple TestFlight. He says that not only how to use tools, but also the eye for choosing tools, is cultivated. He emphasizes the value of learning by watching others at work.

The Author’s Perspective and Aims in

Revising His Book

He says he is currently writing the second edition of “Rust Web Development.” When he brought the proposal to the publisher Manning, he asked himself what it means to write a book in this era. He says that even while writing the first edition, the faces of the people who raised him came to mind. He says he felt the significance of leaving behind an approach to work, beyond just explaining technology. A record bearing the author’s name clarifies where responsibility for answers lies. Readers can judge trust not only by the content, but by looking at the author’s background.

The book form is well suited to visualizing lineage. Revisions from the first edition, acknowledgments, and references accumulate. If there are errors, they are corrected in errata and the next edition. In that trust in infrastructure depends on provenance and transparency of management, this overlaps with the lesson shown in Microsoft Secure Boot Left Vulnerable for 13 Years. Who has maintained it and how it has been updated is called into question. He argues that there lies a temporal responsibility absent from the fluent summaries of language models.

Implications and Challenges for Talent

Development in the Field

This essay is also relevant to designing training in the workplace. Rather than banning language models, their positioning needs to be clarified. A possible division of labor is to use machines for initial understanding and multifaceted explanations, while confirming the basis for judgments with people. He says that in areas involving responsibility, such as design decisions and incident response, the presence of experienced colleagues is indispensable. Apprentice-like collegial relationships need to be deliberately preserved. He says instruction combining documents and dialogue is effective.

There are also implications for evaluation systems. Beyond speed of output, it becomes important to record who inherited what from whom. If the names of advisers and the versions of materials are left behind, they can be verified later. Reasons for technology choices also remain as history. He says that an organization’s memory is a bundle of individual memories. He argues that deliberate investment is needed so that mentoring relationships do not wither behind efficiency gains.

Editorial Opinion

In the short term, we expect to see verification procedures developed in workplaces that use generative AI for learning support. Practices requiring the source of answers and version control may spread. Attempts to preserve apprentice-like dialogue among colleagues are also likely to increase.

In the long term, we assess that standards for talent development will shift to disclosure of lineage and track record. Books and courses may be chosen based on the author’s background and revision history. We assess that organizations will come to treat documented mentoring relationships as assets.

The remaining question may be how far disclosure of who one learned from should go. How to distinguish statistical prediction from human responsibility is at issue. We see identifying the conditions for balancing efficiency and trust as the challenge.

References

• “It matters who teaches you”, by bastiangruber.ca by gruberb — Lobsters, 2026-09-05T17:47:34.000Z (ARR)
• Source URL: https://bastiangruber.ca/posts/it-matters-who-teaches-you/