Hacker News Reader

Comment by dave_sullivan | Hacker News Reader

If you read an article by an algorithm and then discuss it as if it were written by a human, was the turning test passed?

View on HN

The article is not written by an AI - all the facts are correct, and the internal references within the sentences are all correct. I have used invisible iframes as security sandboxes for third party code. The “Modular Development” section is weak IMHO, since visible iframes introduce UI and UX issues.

It reads like something a web developer might write. I googled two sentences as a test, and they were both unique to that page and site.

Edit: code quality is questionable - by my reckoning. Look at code buttons on this page: https://celody.com/modules.html

> I googled two sentences as a test, and they were both unique to that page and site

It's not impossible to generate new sentences with AI, we've been able to do this for a long time.

The footer on https://stakedy.com/long/how-staking-on-stakedy-works.html (the about page) says:

> Stakedy is a product from digital people within Celody

I'm not sure if that means that the Authors of the posts are AIs, or if it's just made from the same people that made Celody. Maybe the ambiguity is on purpose?

> The article is not written by an AI - all the facts are correct, and the internal references within the sentences are all correct.

I don't see how this follows.

You seem to have started from the position that an AI can not write an article with correct facts and internal consistency, and then argued that since this article has correct facts and internal consistency, it can not have been written by an AI?

I am saying the quality of the content of that article is beyond the known capabilities of our current ML systems.

The overview is better than what you might find on stack exchange about using iframes for security sandboxes (Cue joke about intelligence and stack exchange).

Your comment is just saying “if you are responding to something that passes the Turing Test, how can you tell?”.

While the GPT-x kind of AI-text-generators had issues with grammar and internal consistency, they are not the only generators. I know of at least one industry (used cars) where the descriptions of the cars have all been AI-generated for the last 10 years.

Is there some service provider/tool for this that is widely used in the used car industry? Or just convergence on a common approach?

It's basically the service provider thing. Essentially, almost every used car lot in the US uses the same company for their website. They get their own domain and branding and it's mostly a unified database of cars across the country.

And it's not cool. They figured out a lot of dark patterns, like "number of photos, regardless of quality, is a strong driver to sale", so they started duplicating photos in a set to bump up the total photo count. They figured out they could reuse walk-around videos for any same make/model/trim/color across the country and nobody would notice when they walked in the lot the different door panel dings and such. Or they'd display a car that matches your very specific search, that was actually hundreds of miles away, to get you on the lot just to say "damn, we just sold it, but we got this other one in a different color".

I worked for them for about 3 months before I literally couldn't stomach it anymore. I threw up in my driveway, went to work, and gave my notice.

well, most people don't read the article. Only the headline.

Some look at it, but take it as a reason to argue knowingly.

not easy to take a conclusion ...

Reply on news.ycombinator.com