I had a brainwave recently. I was tired and looking at two XML documents which looked identical to me and I thought hey, let's see what ChatGPT thinks. So I asked it to describe to me what the differences were between the two documents and it immediately just started talking about elements that it had completely made up. Every time I asked it why it was doing that it apologised but then doubled-down on making even more stuff up. Eventually I asked it to show me what its understanding was of the two documents I was asking it to compare and it showed me two completely unrelated XML documents

I do not understand why people expect chatgpt to reason when all it is is a fancy probabilistic language model....

I guess humans have high bias towards trusting confident-sounding language despite what reason would tell us to do. That's how politicians and advertisment work anyway....

I just thought it would be interesting, given that it has an understanding of XML to see if it could do a simple diff, "by eye" if you will. Obviously I wasn't intending to trust its output. We of course have long standing trusted tools for diffing files.

Unless your xml input was very small you probably exceeded it's input window.

I suspect chatgpt does some summarization under the covers but otherwise it still only has a finite and fairly small lookback.

I suspect you're right there. Yet it forged ahead confidently giving answers anyway!

Right, but consider its 'evaluation' during training: During training it is constantly seeing stuff where the context is out of the window and the correct completion confidently answers, so the model is trained to do the same.

I think this is very tricky to solve conceptually (since the human authors don't have the same input event horizon problem), but it could be (and has been) papered over by making the context bigger.

I definitely "cheated" some of these by throwing in something horribly verbose rather than whatever a more correct solution would be

I felt dirty when I did something like

    #one, #three, #five, #six, #nine

(Or whatever the elements we had to select were.)

Same here, and to be honest I'm still not sure what the clean solution would be for this level. With that particular html structure and class/id usage there maybe isn't any…

The hint suggests doing it this way. This solution is clean but it isn't very elegant.

Exactly. All the previous questions had real, elegant answers. This one felt like whacking the mole was _possible_, but not the right way.

Europe has entered the chat

Draw.io does this. When you export a diagram as a PNG. There is an option to embed the source file in the png. If you subsequently open one of those PNGs in Draw.io you can carry on editing it. I find it really handy.

Whoa. Had no idea!

Does anyone else remember "Web 2.0" being talked about as "The Semantic Web"? Because I sure do.

It was actually 3.0, but yeah.

[1] https://en.wikipedia.org/wiki/Semantic_Web

I thought that 3.0 would be semantic and the current web was noot semantic.

Is it 1998 again?

Throwback Friday, apparently. I'm almost disappointed that Firefox doesn't Substitut hn orange with the old /. petrol for a page with this content.

Is getting old mostly about watching people rediscover the same mistakes over and over again? I think I am getting old.

This is my exact takeaway. I can't decide whether this article and many of the commenters are deliberately missing this point or whether it's actually not understood.

This seems to miss the point. Of course these dithered images don't result in a saving in filesize when offered up as jpegs.

This feels like a bit of a strawman argument.

Some people here weren't web developers in the 90s and it shows :)

One of the things that gets to me while working from home is the lack of ambient noise. Where I live is fairly rural and the silence can be deafening