Codex, Opus, Gemini try to build Counter Strike

Original Article ↗ Hacker News Discussion ↗

Overall reaction & nostalgia

  • Many found the experiment fun and nostalgic, evoking CS 1.6 / Source era and old-school Quake/Doom aesthetics.
  • Several commenters emphasize that, despite the Counter-Strike framing, the result is really a very simple, Minecraft-looking generic FPS, far from a real CS-like game.
  • Some enjoyed actually playing the demos, mentioning bugs (e.g., farming kills by repeatedly shooting dead players) and getting repeatedly insta-killed.

Technical quality & limitations

  • Multiple commenters say this is roughly what a junior dev could produce in their first weeks: a bare-bones demo glued together from three.js and a backend, with little attention to architecture, netcode, or competitive FPS design.
  • Shooting and networking are implemented naively (send “shot” events and directly reduce HP), which experienced game devs note is nothing like how real competitive shooters handle hit detection, latency, or prediction.
  • Missing features: lobbies, robust physics, proper game modes, cheat prevention, and production-grade engineering; it’s called a “pre-prototype” at best.

Copyright, licensing, and LLM training

  • A shader snippet referencing “Preetham” raised suspicion of LLM plagiarism; investigation shows it originates from three.js examples (MIT-licensed) and/or common implementations of a 1999 daylight model.
  • This sparks a broader debate:
    • Concern that LLMs regurgitate licensed or unlicensed code without notice, creating business risk.
    • Counterarguments that small algorithmic snippets are hard to meaningfully copyright, and this particular case wasn’t LLM output but a bundled dependency.
    • Discussion of derivative works, court rulings on generated output, and fear of copyright trolls versus the practical limits of enforcement.

Impact on developers & workflow

  • Some developers feel depressed that LLMs may remove the “fun” parts of coding, leaving review and bug-chasing in low-quality “it-compiles” codebases.
  • Others say LLMs have made programming more enjoyable and productive, offloading boilerplate, plumbing, and scaffolding so they can focus on design and harder problems.
  • Consensus that LLMs currently behave like junior developers: useful with guidance, but far from autonomous or production-safe.

Economics, usefulness & moving goalposts

  • Skeptics highlight the high cost of complex agentic workflows (e.g., multi-thousand-dollar research tasks) and call these outputs “costly slop” with unclear economics.
  • Supporters give concrete examples where LLMs saved weeks or months (e.g., generating UI mockups, reviving old projects, fixing legacy builds).
  • Several note the rapid “moving goalposts”: what was recently impressive (a crude FPS built from scratch) is now quickly dismissed as trivial or insufficient.

Model comparisons & benchmarks

  • Some claim Gemini’s version is worst and benchmark marketing overstates its real-world performance; others say it actually feels better to play than some alternatives, aside from odd graphics.
  • Discussion of “thinking levels” / parameters leads to debate about whether such knobs are genuine capability or just overcomplication.