AI models may struggle with clear communication, study finds
A new study from New York University and The University of Texas examines why AI-generated text can be harder to understand than expected. Researchers found that modern models like ChatGPT, Claude, and Gemini often prioritize efficiency in problem-solving over human-friendly clarity. This leads to compressed, ambiguous phrasing—such as inanimate subjects acting like living things or stacked noun…
Key points
- 14 AI models tested showed frontier models use more complex, ambiguous phrasing than smaller ones or humans
- Compressed language (e.g., inanimate subjects, stacked nouns) increases reader effort to decode meaning
- Penalizing long responses during training reinforced harder-to-understand sentence structures
The study analyzed 14 models across chat, coding, and essay tasks, comparing their outputs to human-written texts. Frontier models like GPT-5.6 and Claude Opus 5.5 showed stronger tendencies toward complex constructions, while smaller models like Llama and Olmo Base performed closer to human writing. Experiments also demonstrated that fine-tuning models to penalize longer responses reinforced compressed, harder-to-parse language—even for untrained sentence structures. The authors suggest this trend may stem from training objectives that favor efficiency over readability.
Model page: Claude Opus 5.5 →
Why AI-Generated Text Is Harder to Understand Than You Think
Unite.AI · 26 October 2026
Loading the full article…
This text was published by Unite.AI and written by Martin Anderson. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- OpenAI and Anthropic cut AI model prices by up to 80% in four months · 2 src
- OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracy · 2 src
- Altman warns AI shifts cybersecurity from patches to AI-driven defenses · 1 src
- Moonshot AI releases Kimi K3.1 Pro, scores 55 on Artificial Analysis Intelligence Index · 20 src
- OpenAI may release GPT bell with security upgrades · 1 src
Comments
via GitHub Discussions