frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Guess which of these LLM outputs is watermarked

https://sgoedecke.github.io/watermark-quiz/
25•gfysfm•2d ago
https://www.seangoedecke.com/readers-cant-identify-watermark...

Comments

arcwhite•2d ago
Interesting, I did very badly, 3/10!
bastawhiz•6m ago
Same, doing worse than random chance seems like an interesting signal though, but I'm not sure what it's a signal of.
fwlr•2d ago
Utterly imperceptible, even when studied under the microscope in a way that LLM text very rarely is in practice.

It will be interesting to see whose concerns are assuaged (perhaps they genuinely though mistakenly believed it would degrade quality), and whose concerns are heightened (perhaps their real objection is that their AI-generated text will become detectable).

NotPractical•2d ago
Could do with some context on how watermarking works. Objectively speaking it should be impossible to tell.
marcyb5st•15m ago
My understanding is that watermarking in prose is basically a bias when sampling tokens. For a system that knows the average probability for each possible token in the LLM vocabulary it is possbile to quantify said bias given enough text.

For a human that doesn't reason in tokens and therefore doesn't know anything about their probability distribution, it should be impossible to tell. Relying on fancy words/constructs within sentences should not give you any signal as well, since you don't know if the the prompt included instructions for that.

madarcho•2d ago
If SynthID is a google technology, then this is likely just us training their ai again, captcha all over again.
Noumenon72•2d ago
Please report success/failure after each test. Asking me to read and compare 30 writing samples to get any feedback at all means I won't finish. Telling me immediately when I got one wrong lets me recognize patterns and improve my guesses.
dozerly•28m ago
Yea, I did two and then harrumphed in annoyance that I was expected to do all 10.
stranded22•3m ago
Yes. Did one - saw that I wouldn’t get feedback until I have completed all 10 (if at all) and noped out.
elikoga•19h ago
I disliked the fact that the experiment only covered prose, which my eyes glossed over and made me actually do random entries to pass on and see the results. I'd love to see it on a more accurate output distribution like commented code
gjm11•31m ago
I don't think anyone is, or plans to be, watermarking AI-generated code as opposed to text.
red_admiral•13m ago
Why not? It would help with a lot of potential legal issues.
demibabs•11m ago
What do you mean? The watermarking applies to all text outputs, including code.

It’s just much less effective since code is low entropy.

raincole•4m ago
https://www.anthropic.com/news/claude-text-watermark

> code—which in very many cases has to be exact—has generally less watermarking than some other forms of text.

Generally less watermarking. Not no watermarking.

kshmir•40m ago
Thought there were only 2 options!
lacker•18m ago
This is like giving you three outputs from md5sum and asking you to guess for which one the input ended in a "q". There's no way to tell unless you break the RNG.
lolokbro•13m ago
Count the kernels of corn in my spiraling torpedo shit as I blast SaaS out my ass!

Bitch I prompt. While you merely romp.

hdiutil is deprecated in macOS 27 Golden Gate

https://lapcatsoftware.com/articles/2026/8/7.html
28•zdw•26m ago•2 comments

Scrap

https://twitter.com/moxie/status/2091218652133732491
52•tosh•1h ago•9 comments

ElevenLabs, TwelveLabs, ThirteenLabs

https://quantumi.sh/public/labs.html
195•jemoka•4h ago•64 comments

Hister – A private, full content search index that you control

https://hister.org/
75•auraham•3d ago•18 comments

A Friendly Introduction to Racket

https://geometridae.bearblog.dev/a-friendly-introduction-to-racket/
91•signa11•5h ago•30 comments

How a Texas student blew the whistle on a rogue AI hacking attempt

https://www.reuters.com/world/how-texas-student-blew-whistle-rogue-ai-hacking-attempt-2026-08-20/
30•olalonde•1d ago•3 comments

typ.ing

https://typ.ing/
59•bookofjoe•4d ago•21 comments

RF Cafe

https://www.rfcafe.com/
55•gregsadetsky•3d ago•8 comments

NetBSD and My Life (2005)

https://mail-index.netbsd.org/netbsd-advocacy/2005/09/10/0000.html
3•gnyeki•23m ago•0 comments

Munder Difflin – Agent harness to run an office of your clones

https://munderdiffl.in/
222•simonpure•9h ago•92 comments

ATProto spaces: A new extension to ATProto that enables non-public data

https://atproto.com/blog/atproto-spaces-alpha
31•grappler•1d ago•4 comments

Why it might be time to rethink the human family tree

https://nautil.us/why-it-might-be-time-to-rethink-the-human-family-tree-1283985
25•Anon84•2d ago•15 comments

Guess which of these LLM outputs is watermarked

https://sgoedecke.github.io/watermark-quiz/
25•gfysfm•2d ago•19 comments

Z80 – The 1970s Microprocessor Still Alive (2021)

https://www.computer.org/csdl/magazine/mi/2021/06/09623402/1yJTvlRLmhi
96•asdefghyk•9h ago•48 comments

Canada will match US tariffs 'dollar for dollar' as trade talks break down

https://www.bbc.com/news/articles/cvgvyy4x2mvo
213•tartoran•13h ago•831 comments

Mythic's analog compute-in-memory architecture

https://www.mythic.ai
20•janandonly•3d ago•0 comments

New MCP Roadmap

https://blog.modelcontextprotocol.io/posts/mcp-roadmap/
141•pentagrama•5h ago•110 comments

Ameliorate

https://ameliorate.app/
63•hakkikonu•1d ago•18 comments

ProgramBench Vetted: Reverse Engineering from a Runnable Binary

https://vetto.ai/companies/programbench-vetted.html
15•rigelbm•2d ago•1 comments

Chinese robot runs 100M sprint quicker than Usain Bolt's world record

https://www.theguardian.com/sport/2026/aug/22/chinese-robot-runs-100m-sprint-quicker-usain-bolt-w...
46•bookofjoe•5h ago•45 comments

MiniageOS: "Dumbphone" Version of LineageOS

https://github.com/ofdryads/miniageOS
20•ashenke•2d ago•12 comments

One night in Uzbekistan: Why was this one data point so influential?

https://statmodeling.stat.columbia.edu/2026/08/20/we-couldnt-reproduce-their-findings-and-realize...
15•paulpauper•1d ago•2 comments

Show HN: Rotation via Double Reflection

https://static.laszlokorte.de/rotor-reflect/
45•laszlokorte•1d ago•9 comments

Anthropic appears to be A/B testing reduced effort levels in Claude Code

https://twitter.com/argofowl/status/2091150597374537729
70•matthieu_bl•2h ago•79 comments

The Creation of Abulafia

https://blog.veitheller.de/abulafia.html
15•saulpw•22h ago•2 comments

Show HN: Anonymous age verification with passkey-powered encryption

https://loginwithone.com/
24•mikeysight•3d ago•13 comments

What's in a PowerPoint File?

https://editide.com/blog/what-is-a-pptx-file/
23•danielochoa0620•3d ago•18 comments

Belgian car salesman becomes prince after DNA test proves royal parentage

https://www.cnn.com/2026/08/22/europe/prince-belgium-secret-son-scli-intl
58•MilnerRoute•2h ago•41 comments

Show HN: terminal-code – VS Code inside the terminal

https://terminal-code.com
21•robpruzan•3d ago•4 comments

Rust Glancer: Rust LSP using 100x less RAM

https://rust-glancer.github.io/blog/hello-world/
373•matklad•23h ago•87 comments