ethanj8011

Thanks. Looks like it does line up with the original findings, however I do feel pretty confident that this could be ameliorated using slightly more complex watermarking strategies. (I do think that overall watermarking is not an effective governance solution for text)

> Natzir Turrado, a technical consultant, actually tested this (article in Spanish) by reproducing SynthID-Text on an open model with the paper's settings and a watermark key. He asked the watermarked model five times to describe a memorable trip to a coastal town. Four of the five answers he got back opened in a near-identical way: “My heart still sings/dances when I think back to my trip to the quaint coastal town of…”. For the unwatermarked model, three opened in different ways.

A test of 2/5 vs 4/5 seems to be an obviously invalid experiment.

"How we did it" (Claude and him).

Some bad writing examples I liked.

> By the third pause to look up a word, your tea has gone cold.

And here I thought that water had one of the highest specific heat capacities.

> Both numbers are true

What does "both numbers" refer to? Presumably the two percentages, (based on the title, but there is no framing right away that these are the key numbers), and proximate to the line are several unrounded numbers.

> is a flashcard app that teaches Japanese from anime. Each card uses a line where you know everything except the one new word, which is why we started counting lines. The figures above count each word once, however many meanings it has. That wouldn't work in the app: knowing あれ as "that" doesn't help when someone says あれ？ (huh?). These are two separate cards. It doesn't catch every such word yet, we're working on it. During early access it offers one show, Hunter x Hunter.

Highly advise cutting down your marketing pitch for your product, it's just too long to grab people.

Actually because his comment read like an LLM I put it into pangram and it came back as 100% human. Take what you will from this.

Megaslop. I stopped reading at "Your thinking stops disappearing. Your decisions start compounding."

Like what even is this, we know about ADRs, this is a common framework.

> For me, around 200,000 or 300,000 is where I start looking. I’ll check, see it’s at 250,000, and think: I’ve got a big job coming up. I’m going to reset and start fresh

I don't find this to be true, at least with the newer models. I also find that the models fill up to 250k very quickly (presumably mostly by looking through the codebase). Without that context I find the models perform badly. I also don't really notice that large of a change until around 700-800k context and even then it isn't as bad as it used to be. I suspect, that (especially for coding), the long horizon RLVR tasks have helped with this quite a bit.

A combination of words that are infrequently observed in natural language. For example in the preceding sentence some of the 3-grams are "combination of words", "infrequently observed in", and "words that are". N-grams are just this for all values of n. They are saying that often times the exact sequence of a short fragment of text is not very common in human written text.

This was excellent, and one would hope they can at least try to optimize these metrics a bit.

Obviously AI written text. At least there is real content here, however I do find the order and the way in which it was written and portrayed to have a heavy AI smell. That is, it's hard to read and follow the clean narrative, it isn't written the way a human who was trying to convince another human would write, and is less effective because of it.

I don't think this told me anything, and it reads like basically something I could have asked claude to whip up for me.

I hate this website's AI slop design man. Just to pick out some pieces. Why the random circles, why do I care how many people are "here". Why is the text of the number of incidents so small. Why is everything so busy, why is there so much text full of things I don't need to read. Why are they all presented at the same loudness.

I like the idea, but it's a little too obviously AI written for my taste.

slop core

slop core