Hacker News Reader

Comment by Telemakhos | Hacker News Reader

Is this an example of GPT-4 generated text?

Nowadays, I assume that anything with the phrase "it's important to consider" is ChatGPT-generated.

It is important to change my way of writing English then.

Sounds like the metric is, in isolation unreliable and you're going to have a lot of false positives. Better to treat it as an indicator and rely on more nuanced evaluation.

I asked GPT to continue the conversation :)

quantumconfusion 6 hours ago | root | parent | next [–]

True, but there's also the flip side. People might inadvertently adopt AI-like patterns in their writing, especially as AI becomes more pervasive in our daily lives. It's a curious kind of feedback loop.

42Philosophers 6 hours ago | root | parent | next [–]

I think that's already happening. Sometimes I catch myself phrasing things in a way that sounds like an AI response. It's weirdly fascinating and a bit unsettling.

nerd_visionary 6 hours ago | root | parent | next [–]

That's the essence of language evolution, isn't it? We mimic the structures and patterns we're exposed to. In the past, it was literature or media, now it's AI. Just another chapter in the linguistic journey.

digital_scribe 5 hours ago | root | parent | next [–]

I wonder if we'll see a new form of 'AI-influenced' dialect emerging. The impact of technology on language has always been significant, but this could be a whole new level of influence.

All you need is an importance to note

It's important to note that the use of the phrase "its important to consider" is not a reliable indicator that text was generated by ChatGPT. It is a common phrase used by many people across the world.

I think it is, if you look at the user’s comment history it seems to alternate between GPT style and a more natural human HN style.

I’ve noticed this pattern among both HN and Reddit accounts, though for the life of me I can’t understand why.

Surely the person recognizes that their ChatGPT generated replies are nothing like their actual writing? And that it’s very obvious when they’re using prompted text with no custom instructions…

You’d expect a product advertised as intelligent and potentially world ending can do better than a template engine squirting ‘${context} ${solution|answer} ${conclusion|disclaimer}’. Because that how text generators sound like - so easy to detect.

It doesn't really have anything to do with being a "text generator". GPT-4 sounds like a bland robot butler because that's exactly what post-training reinforcement learning make it sound like. It's intentional.

Yes, I work on systems that integrate LLMs and search technologies and frequently test whether or not what is said might be better run through an LLM. I don't post things randomly, however. Sometimes the output is good enough to post, sometimes (clearly) it isn't. However, it's not my default behavior.

Reply on news.ycombinator.com