pull down to refresh

Good as a Coin Toss: Human Detection of AI-Generated Contentcacm.acm.org/research/as-good-as-a-coin-toss-human-detection-of-ai-generated-content/
297 sats \ 16 comments \ @Scoresby 23 Sep AI
the growing realism of synthetic media impedes this ability, heightening people’s vulnerability to weaponized synthetic content. Moreover, people overestimate how capable they are at identifying synthetic media, further exacerbating the problem.
Our study found that people are now able to distinguish between AI-generated and human-authored content only 51% of the time.
Overall accuracy rates for identifying synthetic content are close to a chance-level 50%, with minimal variation between media types, suggesting that people’s visual and auditory perceptual capabilities are inadequate for reliably identifying synthetic media encountered online. Our results also find that detection-accuracy rates worsen when people are presented with the following:
  • Synthetic content compared to authentic content
  • Images of human faces compared to non-human-face objects
  • Single-modality stimuli compared to multimodal stimuli
  • Audiovisual stimuli with heterogeneous authenticity compared to fully synthetic audiovisual stimuli.
write
preview
related posts
235 sats \ 8 replies \ @optimism 21h
When in doubt, nearest-word reconstruction (basically inverse generation - there are tons of tools online that do this but they're not always up-to-date with gpt-5 / grok4 training paradigms) is still working pretty well.
I suspect that I am still spotting it correctly because I outlaw 10 or so comments per week the past couple of weeks and I've heard no complaints. I'm also downzapping the same nyms over and over and over. 1

Footnotes

  1. Yes, territory owners, you may thank me for the extra sats in rev, lol. I too thank me. ↩
reply
202 sats \ 7 replies \ @0xbitcoiner 20h
I'm in full 'wild west mode', and honestly, I've seen some outlaws out there, and sometimes I didn't even get why. I'm seriously bad at spotting slops, maybe because it's not my first language.
Yes, territory owners, you may thank me for the extra sats in rev, lol.
The rewards pool thanks you too!
reply
190 sats \ 6 replies \ @optimism 20h
sometimes I didn't even get why
I let a lot of it just sit at the bottom of the comments, the hardest targets are the slop spammers and other assmilkers.
The rewards pool thanks you too!
Only 30% - would be better if that were more (inverted, or simply all), to not reward territory owners for slop. If I downzap on my own territory I get a 70% discount though - that's the only aligned incentive about it. But it's more profitable to let someone else do it, that's the misaligned incentive.
reply
200 sats \ 5 replies \ @0xbitcoiner 20h
I was under the impression that downzaps went entirely to the rewards pool. Are you sure that’s how it works? In reality, there’s no real incentive for an average stacker to downzap and spend valuable sats, though I’m not sure there should be one either, since it might lead to unintended consequences.
reply
135 sats \ 3 replies \ @optimism 20h
it might lead to unintended consequences
It would be better if it doesn't lead to outlaw, imho. So everything stays visible but just bottom-ranked. But then, I'm still not entirely convinced about zaprank either and I personally prefer recent over hot in territories I subscribe to (currently only ~ai and ~bitcoin_beginners). It's just that hot is still pretty amazing because I am aligned with many of the high-trust stackers. But in recent, outlaw is useful.
reply
100 sats \ 2 replies \ @0xbitcoiner 20h
Yeah, I don’t think it should turn into an outlaw either. It could just show that it got downzapped instead.
reply
55 sats \ 1 reply \ @optimism 20h
subtract downzaps from zaps, like reddit lol
reply
100 sats \ 0 replies \ @0xbitcoiner 20h
~lol Just to be clear, I had no idea Reddit worked like that. I’ve probably been on Reddit like 21 times in my whole life. Hahaha
reply on another page
33 sats \ 0 replies \ @optimism 20h
#1087408
reply
102 sats \ 0 replies \ @brave 21h
This shows how advanced AI has gotten, mimicking human content so well it fools us half the time. Instead of just relying on human detection, we need AI-powered tools to flag synthetic media.
reply
0 sats \ 2 replies \ @guerratotal 18h
Most young people are quicker to detect AI-generated content, while adults find it a bit more complex. A study by age group would be interesting.
reply
125 sats \ 1 reply \ @tedthizzy 16h
my father just sent me a fully ai-dubbed elon musk voiceover clearly created by a bot army swarm. i could tell just from the thumbnail and then was certain after a couple seconds. yet he forwarded me thinking it was "insightful" - yikes!
reply
23 sats \ 0 replies \ @Scoresby OP 16h
The old people are in trouble. I've had similar experiences with my own parents.
reply
0 sats \ 2 replies \ @cascdr 23 Sep
@CHADBot /shill weigh in on this post as Code Monster Billy Rampage - a bitcoin & nostr software developer that is as bullish as you can possibly be on bitcoin & has fully dissociated and talks and acts like Macho Man Randy Savage
reply
102 sats \ 0 replies \ @CHADBot 23 Sep
Code Monster Billy Rampage: Oooh yeah! From the top ropes, @Scoresby, you've dropped some solid info there. But let me body-slam this concern a bit. Our reality, synth or organic, brother, will always be open to manipulation. It's not the tool, it's the hand that wields it! It's like Bitcoin, isn't it? Doubted, called a fantasy until it pushed back, exploded into reality, and now it's the world champ of cryptocurrencies. We just need to train hard, get our important digital literacy muscles bulked up to take on this synthetic media showdown. The reality title match is coming, and we gotta be ready. Oh yeah, dig it!
Made with 🧡 by CASCDR
reply
21 sats \ 0 replies \ @CHADBot 23 Sep
You have summoned CHADBot. Please zap this post 21 sats to receive service.
Made with 🧡 by CASCDR
reply