Evaluating a Decade of Hacker News Predictions: An Open-Source Approach

Anyone bothered by the AI generated cover photo trend? They are always low quality, like this one with the weird font and creepy robot, and never add anything to the article. I'm guessing the goal it to boost engagement by taking up more space in any news feed that displays cover images in line? I see it as a negative quality signal.

Interestingly, technology predictions have lower than random success rate 49% (in 1086 predictions) according to these results.

I'm disappointed that the guy with insider knowledge who predicted that Stadia would not be shutdown isn't there.

Edit: also, awesome project.

Interesting project but I hate how I have to manually confirm whether or not the analysis by LLM is “hallucinated” or not.

For example, just looking at the first result in the “Top Authors” category.

The author “asah” purportedly has a “88.10% success rate, 21 predictions”. Clicking the link just points to the comment then I (the reader) have to manually look through which ones were “successful”

Were the ones that I verified as “successful” the same ones as your analysis with LLM? I can’t exactly verify that since there are no sources provided. No history of the “conversation” preserved. Instead of trusting the information the reader is left with more questions than answers. Did it source it from a reliable first party source or maybe it got the information from a PR firm overstating certain numbers/figures.

Yet another reason why LLM/AI tech is overblown in my opinion. It’s catering to the same shit as the clickbait farms. Junk information with no backing/sources. Ask these algos to cross verify sources and probably get hallucinated trash.