Skip to content
← All problems
#30 by prioritySteadyAI

Outdated knowledge of libraries and APIs

Models use deprecated APIs or old library versions due to training cutoffs.

Priority (½-life 90d)
1.6
Total mentions
6
Distinct people
6
Last 30 days
1
Last 90 vs prior 90
1 / 1
Mentions per monthHover or focus a month for details
024Nov ’252DecJan ’26FebMarAprMayJunJulAugSepOctNov
Show data table
Monthmentions
Nov 20250
Dec 20252
Jan 20261
Feb 20261
Mar 20260
Apr 20260
May 20260
Jun 20261
Jul 20260
Aug 20260
Sep 20260
Oct 20261
Nov 20260

Most often mentioned with

  • Claude 3
  • AI 3
  • LLM 2
  • Git 2
  • Gemini 2
  • Codex 1

What people said · newest first · first seen Dec 5, 2025

  • And this is another example of why you shouldn't let LLM (especially older versions of Claude) to write all game texts. To me it's extremely annoying to read those.

    haruka_ff · Oct 10, 2026 · comment in “Triple-A Minesweeper”

  • Yeah, I can't edit my first post, but should note that if you don't want to rewrite history of the submodule, it also at least lets you consistently do shallow clones of the submodule - various git-related CI/CD operations on the main repo won't work with shallow clones, so it's a pain if you have to pull in a million old versions of binary dependencies just to run gitversion.

    Marsymars · Jun 18, 2026 · comment in “Lore – Open source version control system designed for scalability”

  • Gemini flagged claude-sonnet-4-20250514 as a "fictional model" and issued a critical blocking warning. The model is real — Gemini's training cutoff made it hallucinate about a model name while reviewing a hallucination detector.

    zhangxiaowen · Feb 4, 2026 · story in “Show HN: 32KB deductive engine that catches LLM hallucinations”

  • The second one is that it's a little worse at using tools, though 5.2-codex is pretty good at it. The third is that its knowledge cutoff is further in the past than both Opus 4.5 and Gemini 3 that it's noticeable and annoying when you're working with more recent libraries.

    deaux · Jan 23, 2026 · comment in “I was banned from Claude for scaffolding a Claude.md file?”

  • One of the most frustrating experiences in modern development is the "Knowledge Cutoff." You pick up a bleeding-edge framework—say, the latest alpha of a Rust web server or a brand new JavaScript meta-framework—and your AI assistant hallucinates syntax from three years ago.

    tithos · Dec 10, 2025 · comment in “[dead]”

  • I built CodeContext API (https://codecontext-api-production.up.railway.app) to solve a problem I kept hitting while building AI coding agents: The Knowledge Cutoff. My agents kept hallucinating APIs for libraries like LangChain or Next.js because their training data was stale.

    riskofcollision · Dec 5, 2025 · story in “I built an API to give LLMs instant access to documentation for 1000 libraries”