AI writing still sounds fake due to syntactic patterns, not keywords

Researchers have identified syntactic patterns, rather than specific keywords, as the primary indicator of AI-generated text. An audit of over 134,000 LinkedIn posts revealed that while AI models like GPT-4o and Claude 3.5 Sonnet tend to produce text with consistent sentence lengths (low burstiness), human writing exhibits greater variation. To combat this, a new tool called GRAI has been developed, which acts as a deterministic linter to enforce stylistic diversity and flag common AI-generated phrasing, ensuring a more natural and varied output. AI

IMPACT This research could lead to more sophisticated AI detection tools and encourage the development of LLMs that produce more varied and natural-sounding text.

RANK_REASON The cluster describes research into AI-generated text and the development of a tool to address its perceived flaws. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Claude 3.5 Sonnet
• GPT-4o
• reinforcement learning from human feedback

• product
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →