---
title: "Lingo — GC AI Glossary"
description: "Plain-English definitions of AI lingo such as vibe coding, slop, jailbreaking, stochastic parrot, GPT wrapper, and benchmark hacking."
canonical: "https://gc.ai/glossary/lingo"
last_updated: "2026-11-18"
---

# Lingo — GC AI Glossary

> Plain-English definitions of AI lingo such as vibe coding, slop, jailbreaking, stochastic parrot, GPT wrapper, and benchmark hacking.

- [Agentic looping](https://gc.ai/glossary/lingo#term-agentic-looping): When an AI agent gets stuck calling the same tool or retrying the same step over and over, burning tokens without making real progress.
- [Benchmark hacking](https://gc.ai/glossary/lingo#term-benchmark-hacking): Tuning a model to score well on public evaluation suites at the expense of real-world performance. The AI version of teaching to the test.
- [Context stuffing](https://gc.ai/glossary/lingo#term-context-stuffing): Dumping huge chunks of source material into a prompt instead of retrieving only the relevant parts. Works until the model loses the thread.
- [Doom scrolling the alignment problem](https://gc.ai/glossary/lingo#term-doom-scrolling-alignment): Compulsively reading takes on whether advanced AI will end well for humanity. Usually late at night. Usually on X.
- [Emergent cope](https://gc.ai/glossary/lingo#term-emergent-cope): Insisting that a model's failure mode is actually a sign of intelligence ("it's reasoning about why it shouldn't answer") when really it just hallucinated.
- [Fine-tuning copium](https://gc.ai/glossary/lingo#term-fine-tuning-copium): The persistent belief that any model shortcoming can be solved by fine-tuning on a bigger custom dataset. Usually it can't.
- [GPT wrapper](https://gc.ai/glossary/lingo#term-gpt-wrapper): A dismissive label for a product that's mostly a thin UI over someone else's foundation model. Sometimes deserved, sometimes not.
- [Jailbreaking](https://gc.ai/glossary/lingo#term-jailbreaking): Crafting prompts designed to bypass a model's safety guardrails and coax it into producing content it's been trained to refuse.
- [Model mogging](https://gc.ai/glossary/lingo#term-model-mogging): Showing off that you have access to a newer, more powerful, or otherwise restricted model than the person you're talking to. From "mogging": outshining someone effortlessly.
- [RAG maxxing](https://gc.ai/glossary/lingo#term-rag-maxxing): Aggressively expanding a retrieval-augmented generation pipeline (more documents, more chunks, more rerankers) in pursuit of accuracy. Sometimes works.
- [Shoggoth with a smiley face](https://gc.ai/glossary/lingo#term-shoggoth): A meme depicting a vast, alien, multi-tentacled creature wearing a friendly cartoon mask: the inscrutable base model under the polite chat assistant.
- [Slop](https://gc.ai/glossary/lingo#term-slop): Low-quality, generic AI-generated text or images: fluent, plausible, and instantly recognizable as machine-made. The visual cousin of spam.
- [Stochastic parrot](https://gc.ai/glossary/lingo#term-stochastic-parrot): A skeptic's framing of large language models as systems that statistically remix their training data without genuine understanding. From a 2021 paper by Bender, Gebru, McMillan-Major, and Mitchell.
- [Token maxxing](https://gc.ai/glossary/lingo#term-token-maxxing): Cramming the maximum amount of text into a model's context window (every email, every spec, every PDF) on the theory that more input always equals better output. Often the opposite is true.
- [Tool spam](https://gc.ai/glossary/lingo#term-tool-spam): Giving an agent access to so many tools that it can't pick the right one, or calls half a dozen when one would do.
- [Vibe coding](https://gc.ai/glossary/lingo#term-vibe-coding): Building software by describing what you want to an AI and shipping whatever comes out, without reading the code too carefully. Term coined by Andrej Karpathy.
