news.volyx.in

Various LLM Smells (shvbsle.in)

369 points by speckx · 93 days ago · 294 comments on HN

Article summary

The author noticed that LLM-generated writing and AI-generated websites often exhibit similar patterns and structures, which they term 'ai-smells'. These patterns include the use of certain phrases, sentence structures, and design elements that are overly familiar and generic. The author is not opposed to the use of LLMs for creative tasks, but rather observes that the output can be recognizable and lacking in diversity. This observation is based on the author's experience with using LLMs to polish their math blog writing.

Main themes

  • LLM writing patterns
  • AI-generated content
  • Design and aesthetics
  • Language and style
  • Creative writing and authenticity

What commenters say

  • The use of certain phrases and sentence structures can be a giveaway that content is generated by an LLM, and discussing these 'tells' openly may help to improve the diversity of AI-generated content.
  • Some people find LLM-generated writing to be of poor quality and lacking in creativity, while others argue that it is better than what many humans can produce unassisted.
  • The overuse of certain design elements, such as specific fonts and color schemes, can make AI-generated websites and content feel generic and unoriginal.
  • There is a concern that the widespread use of LLMs could lead to a homogenization of content and a lack of aspiration for quality, as people may settle for 'good enough' rather than striving for excellence.
  • Others argue that LLM-generated content can be useful and acceptable in certain contexts, such as when humans lack the time or skill to produce high-quality content themselves.
  • The quality of LLM-generated code is also a topic of debate, with some people questioning why there is a double standard between accepting AI-generated code but rejecting AI-generated prose.
  • Some commenters suggest that the problem lies not with LLMs themselves, but with the culture of mass production and consumption of low-quality content, which can devalue the importance of human creativity and skill.
  • There is a need to distinguish between the potential benefits of LLMs as a tool for assisting human creativity, and the potential risks of relying too heavily on AI-generated content without critical evaluation or human oversight.