k1m

That's really interesting! Thanks for sharing. Will be trying it soon.

> Having less diversity in answers could mean you're more consistently getting a better/worse/same quality answer.

I wrote the article. I'm trying to show how diversity loss is itself a measure of quality loss in many applications. And as I pointed out in another comment, the paper's authors themselves acknowledge as much in their paper.

> Instead of just re-rolling the answer, why not continue in that thread and say what you don't like?

The re-roll is actually a feature of many chat apps with a dedicated button. So it's something users are familiar with.

But the results are also quite different. The prior answers in your example of asking for x alternatives influence the following answers too much in a way a re-roll doesn't.

I wrote the article. How you define quality is exactly the issue.

> In my opinion, 'sample diversity' and 'sample quality' are not the same thing, and conflating the two concepts leads to poor intuition.

But you're assuming everyone has the same notion of quality as you. As I point out in the piece, even the paper's writers use "quality" to show how watermarking can "lower the overall quality of a system"

> SynthID and the like probably matter most if one is trying to repeatedly sample an LLM in order to infer the probability distribution

Which is another impact I write about in the piece. If you evaluate the output quality of an LLM system in this way, you're impacted more by watermarking.

Worth noting that this is what OpenAI wrote about text watermarking two years ago:

> While it has been highly accurate and even effective against localized tampering, such as paraphrasing, it is less robust against globalized tampering; like using translation systems, rewording with another generative model, or asking the model to insert a special character in between every word and then deleting that character - making it trivial to circumvention by bad actors.

> Another important risk we are weighing is that our research suggests the text watermarking method has the potential to disproportionately impact some groups. For example, it could stigmatize use of AI as a useful writing tool for non-native English speakers.

https://openai.com/index/understanding-the-source-of-what-we...

It reduces diversity, which they don't talk about much. Wrote about it here. https://blog.keyvan.net/p/ai-text-watermarking-and-quality

Agree. I think many people forget that not long ago, HTML markup on many sites was a lot richer than it is today. Making it trivial to produce a good trimmed down markdown version.

The reason it may be more difficult today is because we've lost a lot of that. Some of it because of modern JS frameworks, but some also because publishers simply don't want to make it easy for the useful stuff to be scraped and extracted easily.

I'm not convinced that's changing because of AI agents (it's getting worse in many ways with anti-agent rules). Maybe improving for documentation pages intended for agents. But if it is changing, I think it'd be far easier to improve the HTML and let the agent take care of the rest.

In general I think I just don't like the idea of one URL being able to return different content. Forces me to think about what each system I give that URL to may be sending in content negotiation headers. Would rather the HTML is returned and alternatives listed in HTML head.

But for HTML and Markdown in particular, there's been so much useful work done in the semantic HTML space and microformats, that I don't know why anyone interested in this wouldn't just improve their HTML markup and leave it to the agent to do the rest. Convert to markdown or extract the useful HTML before handing it to model.

Thanks. I posted that comment before they had added support - https://news.ycombinator.com/item?id=48353325 - didn't know situation had changed.

But it's something I think developers should think about if they rely on caching. If it took Cloudflare this long to support this, there may be other systems which still don't.

Link to Roy Fielding comment:

https://lists.w3.org/Archives/Public/ietf-http-wg/2013JanMar...

I agree. I think a lot of people here are assuming that the full HTML retrieved has to go into the LLM eating up tokens. But why wouldn't the agent try to clean up first and remove bloat and convert to markdown itself, before feeding into LLM. Semantic HTML would make that easier.

I agree with Roy Fielding on this:

> It is a bad design trade-off to send a bunch of header fields on every request just to tell the server all of the possible variations of preference held by the user, particularly when there is a very small chance that any of those dimensions are applicable to the target resource. It has been a bad design trade-off ever since the very brief period in 1993-94 when folks didn't know which image format would be usable on all UAs and there was no CSS or javascript to allow for client-side adaptation.

> ...The caching impact of proactive negotiation is far worse than the one extra round trip per site for reactive negotiation, and even that round-trip isn't necessary in formats that support client-side adaptation.

On the caching impact, Simon Willison wrote:

> ...you can’t deploy an application that uses content negotiation via the Accept header behind the Cloudflare CDN — for example serving JSON or HTML for the same URL depending on the incoming Accept header. If you do, Cloudflare may serve cached JSON to an HTML client or vice-versa.

Note: I posted this in another comment with links to those two quotes which I couldn't copy easily now - will add later.

I don't know enough about models that produce images, music, video to know how much human thinking and creativity can shape their output. But with text, the kind of output you get from a simple throwaway prompt like 'write me an essay on X' is vastly different from input you've taken a long time to think through and shape. Yet they'll mostly both be treated the same by people who rely on AI detectors.

My biggest issue with AI text is just that it's annoying to read, so I get put off before I can figure out if the content is actually worth reading. I would love a detector that detects the style of writing I find annoying and warns me, or offers to rewrite it so I can give it a chance at least.

> this is only about machine-generated works that are trying to pass as human works.

But by "machine-generated" don't you mean "text that came out of an AI model"? My argument in the post is that such output tells us very little about how much human input went into the original work. It's all lumped together as AI slop.

That's an interesting idea to try as a browser extension. I think Facebook is particularly egregious here, so I doubt you'd need to run something like this on other sites. But if it worked fast enough on just Facebook, it would be useful.

This reminded me of an old documentary about corporations: "If we look at the corporation as a legal person, it exhibits all the characteristics of a psychopath" https://www.youtube.com/watch?v=s5hEiANG4Uk

I personally have no insight into what they have or haven't dismantled. Just found the article interesting (had never heard of the company). Was responding to the "They seized the wrong domain" claim in the tweet you quoted. I read that and thought they disabled an unrelated domain, which doesn't appear to be the case. As for the main domain, no idea about that. Someone brought that up to the author of the piece, and the bit I quoted was his response. I have no more info beyond that.

NearlyFreeSpeech.NET: https://www.nearlyfreespeech.net/about/

Google shows netnut.com[1] being the same thing, so not sure it's the wrong domain. Maybe not the main one. Brian Krebs' comment on this[2]:

> Pretty sure they're working on getting that domain too, but it's taking longer for some reason. Either way, they have dismantled the botnet's back end infrastructure, so for now it's not doing much.

[1] https://www.google.com/search?q=site%3Anetnut.com

[2] https://www.linkedin.com/posts/bkrebs_new-breaking-fbi-seize...

This article, as with many like it, ignores the fact that trust in "traditional media" is at an all time low globally. That's not because people enjoy fake news on social media, it's because traditional media has failed them. There are many studies that look at bias in traditional media. But journalists within that system completely ignore that aspect (it's harder to deal with), and simply assume they just need to embrace videos or new formats.

> I think many people in tech have existed in a high-demand labor market for so long that they have no idea how much less other job markets reward people for their intelligence, work ethic, knowledge, education, etc.

Absolutely! It's fascinating that when I discuss the job issue, and how undemocratic and soul-crushing it is for many people, there's this huge chasm between what most workers experience in their jobs, and that small, vocal group of highly paid, in-demand workers, who enjoy a lot of freedom and who've never experienced the other side.

> From the very inception of these LLM coding tools, some people predicted this would destroy the dev job market, and others said ‘maybe it will replace other software developers but there’s no way it could replace me,’ seemingly assuming their role would remain largely unchanged, regardless of how the field changed.

It's fascinating to watch, and it prickles people's egos. If the thing I spent a lot of time learning to do well, and which gave me status and a high salary, can be done easily by a machine, does that mean I'm not special? I saw this tweet from a very experienced developer, Dax Raad, that captures a part of it well I think[1]:

> i used to say programming was creative work

> except LLMs are fine at programming and are literal 0s for more obviously creative work

> i think we mistook enumerating a lot of possibilities and picking one for being creative

[1] https://x.com/thdxr/status/2064705679298896105

> The fundamental flaw in all these style arguments against core employment is people's personal desire to have a straight forward path for survival that doesn't involve a lot of strategic thought.

One of the most popular alternatives proposed is Universal Basic Income, where you just get a sum of money periodically. That's more straight forward for survival than holding down a job, and worrying about layoffs.

I'm not in the camp that believes AI is actually going to reach human intelligence. So yes, AI is simply dumb. But many people overlook the fact that many of our jobs today require exactly that level of dumbness. So saying AI is dumb doesn't say much about whether it can replace a lot of jobs or not.

My post focused more on trying to question why we value jobs so much, beyond the paychecks they offer. I didn't get into alternatives, but Universal Basic Income (UBI) is what you'll see discussed most often as an alternative. I'm certainly not arguing for taking away people's paychecks and then letting them fend for themselves with nothing to replace it.

On your first point, I agree with you that it's a deeply held belief. But that doesn't make it reasonable. That's why I think it's worth the effort trying to question it, especially at a time when there's so much discussion and fear around job loss.

Regarding UBI, I've also read a lot of different opinions on it, and all the ways it can be implemented to do more harm than good. My own view is that the only reason we even see it proposed from the very wealthy is that they're looking ahead at the direction all this is heading and see it as inevitable (to stave off riots and mass protests that could affect their wealth more than UBI would), so they want their version of UBI to be the one that wins out. That doesn't make UBI as a concept a bad idea, it just means it's going to be a struggle to get a version of it that actually lets us live a decent life. But to even get to that point in the discussion, we have to win more people over to the idea that not having a job isn't such a bad thing, and doesn't say anything about your worth as a human being.

> There is no future in which they willingly submit to the kind of redistribution that would be required to finance a society where work is optional.

Do they ever willingly submit to anything that results in less money for them? Too early to know if AI will actually reduce the need for jobs, but if it does, are they going to be forced to create positions for roles they don't need, so that there are no job losses? And what will people do in those roles?

> because jobs are fucking important and necessary

Why are they important and necessary, and for whom?

I agree, but I think it's a shame that this purpose to life is tied to jobs. It's understandable, because pretty much everyone's path in life is schooling, then a job. So the job becomes our purpose and identity. Job loss can then create a void that people don't know how to fill, because it's all they've had.

Believe it or not, many people genuinely don't see an alternative to having a job. So when AI threatens jobs, rather than call for measures to ensure people can live without jobs, they call for jobs to be protected.

> Without an incentive (i.e., paycheck) for people to work (e.g., code, write articles, answer forum questions, etc.)

Many people do all those things in their spare time, unpaid, because they genuinely enjoy it. Let's not conflate work with a job. https://seths.blog/2026/05/doing-your-job-vs-doing-your-work...

I understand that, and that's what I was trying to get at with the post. That an AI future which can take away jobs, doesn't have to be a bad one if the paycheck is replaced with something else. The Rutger Bergman piece I linked to argues for Universal Basic Income. Many people assume jobs are the only way.

Definition of conflict of interest: a conflict between the private interests and the official responsibilities of a person in a position of trust.

Army generals are in a position of trust, we assume they're acting in the public interest. Whether that's for more spending, more war, or less.

Here's an ex general's view on war and military spending:

> Every gun that is made, every warship launched, every rocket fired signifies, in the final sense, a theft from those who hunger and are not fed, those who are cold and are not clothed. - Eisenhower

The point is, there's clearly a problem if the public are getting opinions on this stuff from people who now have undisclosed private interests.

Shouldn't the public know that the former army general telling them that military spending needs to increase, now has the following roles?

> Paid positions included working part-time as a strategic advisor for Schroders bank, plus advisory roles at Helsing – a German AI defence start-up – and an insurance firm. ...spends 30 days per year “as a thought partner for Tony Blair in his role as Executive Chairman” at the Tony Blair Institute for Global Change. ...is chairman of Equilibrium Gulf Limited, which advises the crown prince of Bahrain on the autocratic country’s notoriously brutal interior ministry.”