It was given today's front page to riff on. Thats why it not only reads like a HN front page, but also has near duplicates from todays front page.

You haven't used Gemini 3.0 Pro if so, this is very much in its range of abilities.

I would think GPT4 probably could have done this.

The first React tutorial I ever did was making a hacker news clone.

It is incredibly well represented in the training data. I am just surprised how many people here obviously must barely use the models to be either overly impressed by this or think it is fake.

Gemini can do WAY more than this.

I can believe. But such texts are not patches to a crufty legacy codebase, they are fun to craft by hand! :)

i agree. the thing that stands out to me is that almost every entry is quite notable - i.e. each entry would have a vote/comment count in at least the multiple-hundreds if not more. and all those events in one homepage seems very unrealistic.

it seems human generated to me or at least prompted with a stronger bias for choosing events that would be interesting to readers, not just a pure extrapolation of the current homepage into the future

This is just absurd.

You can just tell in this thread who really hasn't used these models much at all.

what's absurd? i haven't claimed it's not capable of this, just that the predictions seem more tailored to elicit a certain recognition from humans rather than chosen based on an honest attempt to predict a possible future HN homepage. Since the prompt apparently didn't include such a modifier, and as far as I know Gemini isn't trained by default to do so, I call shenanigans.

Ah yes, the Kasparov approach