HN turned an AI-skeptical essay into a fight about trust, work, and refusal
Original: The future of everything is lies, I guess: Where do we go from here? View original →
HN gave this thread more than 700 comments because Kyle Kingsbury's essay is not a simple rejection of useful tools. It asks what happens when LLMs become part of the default shape of work, education, search, moderation, and software maintenance. The comparison is not to a faster car; it is to the way cars remade cities around them.
The essay's force comes from concrete fatigue. It points to slop in search, hallucinated customer support, machine-written legal errors, LLM-generated pull requests, data-center costs, scraping pressure on small websites, synthetic abuse material, and workers being pushed to outsource judgment. Kingsbury does not argue that LLMs are never useful. The sharper claim is that useful technologies can still impose broad costs when institutions adopt them without accountability.
The HN discussion was tense because many readers recognized both sides. Some argued that technologists who see the risks should stay close to the field so they can say no to the worst deployments. Others were less optimistic, saying the incentives behind large-scale AI are too strong for individual restraint to matter much. Several commenters moved the argument into education, warning that students who skip the painful parts of debugging and writing may lose the practice that builds durable understanding.
The useful community angle is narrower than "use AI" or "reject AI." It is about where verification is possible, where the blast radius is low, and where human responsibility should not be replaced by generated text. HN cared because this is no longer abstract. Developers are already reviewing synthetic patches, debugging AI-written code, handling scraped sites, and explaining why a fluent answer is not the same as a reliable one.
That framing is useful for readers because it separates convenience from accountability. Search quality, code review, education, customer support, and moderation all fail in different ways when generated text is treated as finished work. The thread kept returning to that distinction: productivity gains are real, but so are the systems that quietly transfer responsibility away from people who can be questioned.
Related Articles
The HN discussion focused less on whether AI feels impressive and more on whether the infrastructure math can keep working. Ed Zitron’s essay frames the slowdown question as a financing and revenue problem.
Defense AI startup Anduril Industries raised $5B in a Series H co-led by Thrive Capital and a16z, doubling its valuation to $61B. The company doubled revenue to $2.2B in 2025 and will fund Arsenal-1 manufacturing expansion and Lattice platform advancement.
A video believed to be from Google's unreleased 'Omni' video generation model has leaked, drawing 1,300+ upvotes on r/singularity. Users particularly noted the model's unusually coherent text rendering - a persistent weakness in current video generation models.