Hacker News Reader

Comment by DougN7 | Hacker News Reader

Maybe I’m too naive but I can never tell when something is written by AI. If it works with next most likely token, doesn’t that mean it has encountered the patterns you’re picking out in lots and lots of text written by humans? Please educate me if I’m wrong.

View on HN

> it has encountered the patterns you’re picking out in lots and lots of text written by humans?

In pre-training data, yes

There are post-training datasets, where the weights are changed to conform to human preference. These datasets are created by groups of thousands of people all following a 40-page guide, and these guides have example. People over-index on these examples and so sample sentences with these structures are over represented in these datasets and used for post-training.

Same. Feels like “AI slop” was trained on my personal writing style. The quoted text from the article writes with the same voice as mine.

Same. Everyone wants to feel smart by trying to point out that every piece of writing is AI generated now, but most of us (myself included) are just average writers. All of the LLMs generate phrasing I often use.

Average isn’t median. LLM writes like it has one testicle.

Lol - I’m not even sure what that means.

Okay, it means that since men have two and women have zero, the average person has one testicle. But if you use “average” as a meaningful guideline, you’re going to have trouble because very few people have one testicle; nearly all have two or zero. Here I am making commentary on the quality of AI writing with an analogy to how AI writes like the average person.

Reply on news.ycombinator.com