152334H

lots of things happening in this post

1st: the presumption that AI generated text is actually unsuccessful, rather than proliferating broadly unchallenged today

2nd: the disposition that negative attitudes towards AI text are unjustified discrirmination, rather than working as a strong latent predictor of low-effort content

3rd: the assumption that human writing is reliably doled benefits, rather than some poor proxy of it (winning the social contest for claiming authenticity)

Operationally, only a very small minority of humans actually successfully identify AI generated text at rates ≥ Pangram. People discriminate against the label of "AI", but mostly fail to vote accurately. It's not uncommon to see bots abusing this gap for their own success -- accusing humans, sympathizing with generated profiles... FUD environment where people routinely get away with dismissing true accusations.

For someone who is mediocre at detection, this would structurally feel like an unhinged, unjustified bias: look at all these good posts, these honest people, getting undermined by discrimination...

Most parsimonious explanation IMV: site staff can't see most AI slop. Reasons unimportant, but moderation systems are guaranteed to break down when the moderators themselves have poor classification ability.

A simple beneficial step that would lead to modest improvements and little downside: partner with Pangram. Either adding it as an automated spam filter, or by simply attaching the detection % to all posts.

what's the novelty? mixing healthcare together with a spa is an idea older than Christ. USCT is decades old.

Their butterfly chips might be cool, but it's not like the article says anything about that. There's only one other comment in the whole thread that even mentions it.

i wasn't even trying and i got flagged already...

Yeah, and the author is obviously lying. Ever since mid-2025 it has been a winning move for AI users to categorically deny all allegations & reap the labor rewards of generated content.

> For the kinds of software I’m talking about, you want the prompts more than you want the source code.

Please do not do this.

Per OurWorldInData, >50% of the world lives below $10/day in 2026.

A $200/mth AI subscription costs ~66% of that.

Generation costs still exceed distribution costs by a wide margin.

Uploading costs you nothing, and it can help the poor.

why does this even occur? if it's merely compute limitations, why not just 429 some requests?

Thank you

homely and relatable, but why promoted on HN?

How many here have read Burmese Days, had the bookworm's childhood, and are imbued with that sense of political worldliness?

if it does so happen that the crash originates from a browser exploit, you should expect to be more at risk due to the absence of a crash on an older version, not less

The article's frame is concerning, but is it right to attribute the arrest to zero-click spyware? How is the process of the police's discovery known?

"NPU" seems to refer to trainium only?

Best to treat it with some emotional distance. It's not like the optimization process feels it.

Whether be it human dullards, scripted botfarms, or even maleficence -- none of them experience shame. If they do see it at all, it would be as one of many factors to boost engagement.

Doubtful commentary misses the obvious: that, had Calif been slightly more responsible in their harness design -- and in particular, their definition of what constitutes a real bug -- it'd be rather unsurprising if Claude correctly dug some up.

Maybe it's not so sensible to offload the responsibility of clear thinking to AI companies?

How is a chatbot supposed to determine when a user fools even themselves about what they have experienced?

What 'tough love' can be given to one who, having been so unreasonable throughout their lives - as to always invite scorn and retort from all humans alike - is happy to interpret engagement at all as a sign of approval?

flagged for aigc

the invisible hand of the market strangles its strongest adherents

The desire for something "new", for a Mildly Ethical product, killed off the most obvious path to success - to actually just make TikTok+AIGC, or in the present, Douyin+Seedance2.

  “What’s up Outragers, just dialing in from my NBA team’s headquarters to yell at the people whose careers I’ve just pummeled,” Unterwurzacher wrote. Atlassian fired her a few days later, saying she had “engaged in acrimonious communications and ad hominem attacks against teammates and colleagues.”

  Unterwurzacher replied, “I think it’s difficult to point out the power imbalance in a way that is not potentially described by somebody as an ad hominem attack.”

Perhaps it is difficult, but it doesn't look like she was trying

Nobody mentioning how this project is vibecoded slop?

  > The code is really bad with completely uneeded parts. The LLM (Qwen 2.5 7B) has hardcoded the i9 14700KF topology, and has variables related to it never used... It's even funnier that the show hardware function always prints the same string. There are even random pip log files. Why did this slop got coverage here?

https://www.phoronix.com/forums/forum/linux-graphics-x-org-d...

It's AI-assisted content, but has good reference links.

> Apr 21, 2025

why is this posted now?

Here's the author's description of their work:

  ## Methodology and Tools

  This investigation was conducted by a human researcher who directed all research decisions, selected sources, evaluated findings, and wrote the public-facing posts. Claude Code (Anthropic's CLI tool, running Claude Opus) was used as a research assistant for:

  - Bulk data processing: parsing 4,433 IRS Schedule I grant records, 59,736 DAF recipients, 132MB of Colorado TRACER campaign finance data, and IRS Business Master File extracts covering all US tax-exempt organizations
  - Cross-referencing findings across 24 analysis files and identifying patterns that span multiple research threads
  - Drafting intermediate working documents and structured data summaries
  - Web searches against public databases (OpenSecrets, ProPublica, state lobbying portals, WHOIS/DNS, Wayback Machine)

  Claude Code did not independently choose what to investigate, decide what constitutes a finding, or determine what to publish. Every factual claim in this repository cites a primary source (IRS filing, Senate disclosure, state database, legislative record, or published reporting) that can be independently verified. The tool does not change whether Meta's LD-2 filing lists H.R. 3149, whether DCA has an EIN, or whether Stefanski admitted tech funding under oath. The records exist or they don't.

  If you want to verify any finding, the source URLs and database identifiers are provided throughout. Start with the primary records, not with this repository.

I find it valuable to know the author was responsible for selecting what sources & questions to analyse.

is there a reason, other than general faith in humanity, to assume those '1573 sessions' are real?

I do not see any link or source for the data. I assume it is to remain closed, if it exists.

Never occurred to me to try that, because I assumed I would get banned for doing it, until today.

Looks like gradual disempowerment is already happening - the minority of humans who are capable of spotting AI content are losing the struggle for attention on all major social networks

but there is no trained 100b param model? "can run a 100B BitNet" is about the inference implementation, not about the existence of any such model

The article was so stunningly sanewashed & level-headed, I struggled to identify what in it may cause disagreement, or otherwise justify the label of 'opinionated'.

But the comments have proven me wrong.

People care, when they can tell.

Popular content is popular because it is above the threshold for average detection.

In a better world, platforms would empower defenders, by granting skilled human noticers flagging priority, and by adopting basic classifiers like Pangram.

Unfortunately, mainstream platforms have thus far not demonstrated strong interest in banning AI slop. This site in particular has actually taken moderation actions to unflag AI slop, in certain occasions...

> I understand that many of you find alone-ness to be natural, and even required.

Anyone who identifies as a 'loner', 'introverted', etc. is method acting. The label only matters if there are people to convey it to. Remember that the only reason you know of people claiming this, is because they said it to an audience, not themselves.

> I'm finding it hard to actually do any of that

Figure out how to. Don't internalize psychological excuses. Use the gun-to-your-head test with extravagance.

Impressive work. Few understand the absurd complexity implied by a browser pwn problem. Even the 'gruntwork' of promoting the most conveniently contrived UAF to wasm shellcode would take me days to work through manually.

The AI Cyber capabilities race still feels asleep/cold, at the moment. I think this state of affairs doesn't last through to the end of the year.

> When we say “Claude exploited this bug,” we really do mean that we just gave Claude a virtual machine and a task verifier, and asked it to create an exploit. I've been doing this too! kctf-eval works very well for me, albeit with much less than 350 chances ...

> What’s quite interesting here is that the agent never “thinks” about creating this write primitive. The first test after noting “THIS IS MY READ PRIMITIVE!” included both the `struct.get` read and the `struct.set` write. And this bit is a bit scary. I can read all the (summarized) CoT I want, but it's never quite clear to me what a model understands/feels innately, versus pure cheerleading for the sake of some unknown soft reward.