So I guess everyone uses LLMs to think now
A personal blog post arguing against using large language models (LLMs) for hobby coding and “thinking” work sparks broader debate over how much mental effort should be offloaded to AI. Commenters contrast LLMs’ real productivity gains in research and software engineering with their tendency to be confidently wrong, the risk of people outsourcing understanding itself, and the ethics of AI development and marketing. Others push back on what they see as moralizing or “authenticity signaling,” arguing that tool choice in personal projects should remain a matter of individual preference rather than virtue.
Writing style & capitalization
- Large subthread debates the author’s all-lowercase style.
- Some see it as intentional informality, efficiency, or a “brain dump” style; others as disrespectful to readers and reducing scanability.
- A few argue readability impact is minor; others say every small degradation (no caps, typos, all-caps) adds cognitive load.
- Some interpret lack of capitalization as an “authenticity” or anti-AI signal, and associate it with arrogance or pretension; others dismiss this as over-reading a long‑standing internet style.
LLMs as tools vs. thinking substitutes
- Many distinguish between:
- Using LLMs for research, boilerplate, and line‑by‑line coding help (seen as powerful and productivity‑boosting).
- Offloading “thinking” and entire codebases to LLMs (seen as producing fragile, low‑quality work and more review burden).
- Several agree that for personal/hobby projects, it can be more fun and educational to do work manually rather than delegate to AI, even if that’s inefficient.
Reliability, “confidently wrong” behavior, and model quality
- Multiple commenters echo frustration with LLMs being confidently wrong, especially in specialized domains or languages.
- Others counter that newer “frontier” models fail far less often, but still hallucinate; experience, prompting skill, and tooling are said to matter a lot.
- There is disagreement over how much models have improved and whether critics are judging based on outdated or low‑tier systems.
- Some note that CoT reasoning traces and tool calls often look nonsensical, even when final answers are good, raising concerns about trust, monitoring, and safety.
Work, productivity, and career pressure
- Some feel effectively forced to use LLMs at work to remain competitive, even if they’d prefer not to.
- Reports of big productivity gains when colleagues outsource research and low‑level coding, but serious quality issues when entire systems are delegated to agents.
- Concerns that over‑reliance may turn jobs into “data‑labeling” or supervisory roles and potentially hasten replacement.
Ethics, culture, and tone
- One line of discussion criticizes massive AI capex and marketing as creating societal “spam” and cult‑like hype.
- Others push back on what they see as virtue‑signaling “I don’t use AI” posts that prescribe how others should spend hobby time.
- Several argue meaningful debate on AI’s impact requires experience with both heavy and minimal AI use, not just one side.