Truth is, there are no longer any dead giveaways, let alone any where you can really catch an AI red-handed.
Chat-GPT had, and still has, its quirks: Delves, "underscoring __" (and variants thereof, like "highlighting ___") "it's not just __, it's __," em-dashes, and various characteristic structural and word choices. (It could hardly ever resist ending its responses with a summary paragraph.)
But some of these have been patched out. (I don't think I've seen "delve" in more than a year!) And, especially in GPT-5, the others have become less common and less obvious than they used to be.
Besides, DeepSeek and Kimi-2 write in a completely different and more natural style. Gemini 2.5 is also a very natural writer with a generic style that has fewer identifying characteristics.
So it has become very difficult to identify AI for certain...
codingdave · 4h ago
No. LLMs were trained on human communication. LLMs may over-do some things, but none of those things are unique to LLMs. Every single thing you point at to ID an LLM is also done by humans.
Truth is, there are no longer any dead giveaways, let alone any where you can really catch an AI red-handed.
Chat-GPT had, and still has, its quirks: Delves, "underscoring __" (and variants thereof, like "highlighting ___") "it's not just __, it's __," em-dashes, and various characteristic structural and word choices. (It could hardly ever resist ending its responses with a summary paragraph.)
But some of these have been patched out. (I don't think I've seen "delve" in more than a year!) And, especially in GPT-5, the others have become less common and less obvious than they used to be.
Besides, DeepSeek and Kimi-2 write in a completely different and more natural style. Gemini 2.5 is also a very natural writer with a generic style that has fewer identifying characteristics.
So it has become very difficult to identify AI for certain...