Conversation
For fun I've asked llama3.1:8B to evaluate my 6.12 pull request summary how likely it was written by AI (LLM, ChatGPT). It's a weaker model but I'm still not sure if I should be mad because of "I would assign a probability that it was written by an AI as 95%", or pleased by "I wouldn't rule out the possibility of a skilled human developer having written it.".

Anyway, llama3.1:70B says it's 20% written by AI. Both don't like formal tone and enumerated lists, while acknowledging technical and specific context of linux kernel where AI can "struggle".
1
1
3
The Command-R+ (104B) gives a good answer, 10% and an explanation. Quoted in full below.
Show content
"10%. The text provided contains highly technical content and specialized terminology related to software development and file systems. It discusses specific changes made in a codebase or project log, including performance optimizations, lock scope adjustments, API conversions, error handling improvements, thread management, and various core modifications. This level of detail suggests that the writer has expertise in low-level programming concepts and is familiar with the inner workings of operating system processes.

While AI language models have made significant advancements in generating human-like text on a variety of topics, they typically struggle to produce content requiring deep domain knowledge or highly technical understanding without specific training data related to these complex subjects. The depth of software development expertise displayed here would be challenging for an LLM like ChatGPT unless specifically trained and fine-tuned with extensive data from the field of file systems and low-level programming.

Therefore, it is unlikely that this text was generated solely by AI without significant human input or a highly specialized training dataset tailored to such technical content generation tasks in software development domains."
0
0
1