AI Text Models
The language model behind the fake quotes in our Famous Quotes quiz.
Claude
by Anthropic • claude.ai
Claude is Anthropic's large language model (LLM) and the model used to generate all fake quotes in this quiz. It was trained on a vast corpus of text — books, articles, speeches, and historical documents — giving it a deep understanding of how different thinkers write, argue, and phrase ideas.
How Claude Generates Convincing Fake Quotes
Large language models like Claude don't just produce random text — they learn stylistic patterns from billions of real sentences. For a figure like Einstein, it recognises that authentic quotes tend to be short, metaphorical, and contrast an unexpected pair of concepts. For Churchill, it favours balanced rhetorical structures and dramatic cadence.
The result is fake quotes that feel plausible at a glance: they match the person's known themes, use era-appropriate vocabulary, and carry the right emotional register. That's what makes the quiz genuinely challenging.
The key difference from real quotes is often specificity. Real famous quotes tend to be oddly specific, slightly paradoxical, or rooted in a concrete lived experience. AI-generated alternatives often sound polished but generic — wise-sounding without the rough edges of an actual human thought.
Why AI Text Is Harder to Detect Than AI Images
AI-generated images still have tells: wrong finger counts, blurry backgrounds, inconsistent lighting. But AI-generated text has no physical artifacts — there's nothing equivalent to a malformed hand. The only signals are semantic: word choice, rhythm, tone, authenticity of thought.
This makes Claude's fake quotes particularly hard to spot. Unlike early chatbots that produced obviously wooden output, modern LLMs write with fluency that's indistinguishable from human prose to most readers.
How to Spot AI-Written Text
Because there are no visual artefacts, spotting AI writing means reading for substance, not surface. These are the signals that most reliably separate a real quote or headline from a machine-written one:
- Polish without specificity — AI text sounds wise and reads smoothly, but rarely names a concrete person, place, number or moment. Real thought is usually anchored to something specific.
- Both-sides hedging — models are trained to be balanced, so they avoid committing to a strong or strange position. Real writers take sides and overstate.
- Tidy symmetry — suspiciously balanced sentence structures ("not X, but Y") and neat rule-of-three lists are a stylistic fingerprint of LLM output.
- Missing rough edges — genuine quotes often contain an odd word, an incomplete idea, or a slightly awkward rhythm. AI smooths these away.
- Generic profundity — a line that could plausibly be attributed to any of ten different thinkers is more likely invented than genuinely from one of them.
Put these to work in the Famous Quotes quiz and the Fake News quiz, where one real line hides among AI-written imitations.
Claude at a Glance
Other Major Language Models
Claude is one of several frontier LLMs capable of generating convincing text. Future quiz content may include fakes from:
GPT-4o / GPT-4.5
OpenAI
OpenAI's flagship model. Extremely fluent and versatile, with strong capabilities for mimicking writing styles and producing long-form text. One of the most widely used LLMs in the world.
Gemini 2.5 Pro
Google DeepMind
Google's multimodal model with strong reasoning and writing capabilities. Well-suited for stylistic text generation due to its large training corpus spanning the breadth of the web.
Llama 3.3 / Llama 4
Meta AI
Meta's open-weight model family. Llama models are freely available for researchers and developers, making them a popular choice for offline and privacy-preserving text generation tasks.
Frequently Asked Questions
Why is AI-written text harder to detect than AI images? ▾
AI images leave physical artefacts — malformed hands, broken reflections, garbled text. AI writing has no such tells. The only signals are semantic: word choice, rhythm, how specific and lived-in the thought feels. Modern language models write with a fluency most readers cannot distinguish from human prose, so you have to judge substance rather than surface.
What are the practical tells of AI-generated text? ▾
Suspicious polish without specificity, both-sides hedging that avoids a real stance, tidy balanced sentence structures, an absence of concrete lived detail, and a wise-but-generic tone. Real writing tends to be oddly specific, occasionally rough, and willing to commit to a strange or paradoxical point.
Which AI model writes the fake quotes in the quiz? ▾
The fake quotes in the Famous Quotes quiz are generated with Claude, Anthropic's large language model. Other frontier models such as GPT, Gemini and Llama are equally capable of style-mimicking text and may appear in future quiz content.
Can these models imitate a specific person's writing style? ▾
Yes. Trained on a vast corpus that includes a figure's real writing and speeches, a model learns their themes, cadence and vocabulary well enough to produce plausible new sentences in that voice — which is exactly what makes spotting the real quote genuinely difficult.