Appreciate the feedback on my writing. I never let Claude add its own opinions or interpretations to what I write, but I do use it for improving the clarity of what I want to say. Problem is that I enjoy the way Claude (Opus 4.6, Fable) writes and I'm unintentionally starting to adopt its style so the line is starting to blur.
I get it, it's easy to ask it to just touch up the language and split run ons. At this point in time, I'm of the opinion that "improving the clarity" at the cost of creating "a discernible style" is not a good trade off.
In most formal publications of The Before Times like newspapers and magazines, you did not want much of a writer's specific tone to come thru bc the organization had its own writing style. We used to value "correctness" because too much variance (without intent) made things hard to read—it was like context switching and came with a cost.
Nowadays though, so many people are already reading Claude writing day in and day out. I would venture to say that the opposite cost is being incurred now, where as soon as I start to see the same language patterns, some attention mechanism in my brain just starts to drift off. I literally cannot mentally focus on reading when the prose is 1:1 the same distribution as Claude. Some kind of fatigue happened by reading its output over and over day after day at work.
Anyway, just think about changing it up if you're interested in writing (even if it means your output is "sloppier").
Thanks for flagging this. From what I can tell, it's Beehiiv's locale detection overriding my publication settings, nothing I can change but will file a support ticket with them.
So I understand that the title is a bit inflammatory, but there's a whole section on exactly this, "Am I the Karen Now?". I'm never rude to a rep - they're just following a script to get paid. However, I don't extend any courtesy to the company's AI support bots optimized for stalling and maximizing profits.
Having being on the receiving end of many walls of AI-generated slop. I stand by every message I send, read it thoroughly, and keep it short and precise as to not add to the slop pile.
That's fair. I respect that you scrutinize everything you send and aren't needlessly hostile. I'm less optimistic that most other people who realize they can do this will do the same, and I worry that we'll soon be incentivizing companies to abandon customer service entirely.
Granted, the total is a bit of a reach as "disputes". The bulk of it was IRS + insurance claims. The rest were companies not giving me what I very clearly paid for.
A year with only a $3 problem in it sounds nice, honestly.
Author here. For seven months I've been handing every consumer dispute to a Claude agent instead of eating the cost like I normally would. It builds the case file, pulls the warranty or policy text and the terms of service, works out who actually regulates the company, and plans a "campaign" with fallbacks. Ten unrelated fights (IRS penalties, a car warranty, two insurance denials, a robot mop, grocery stuff), yielded just over $12k back! The biggest single one was an IRS first-time penalty abatement I'd never heard of, $6.8k, eight years into an installment plan.
To be honest, the agent's first-pass advice is usually bad ("file an FTC complaint"); what works is multi-prong approach until something moves. For the robo-mop it ended up all the way at their own arbitration clause, whose consumer fee rules cost them several times the refund. For the pet insurer it was the state regulator plus two exec emails, and they relented in hours after two months of stonewalling.
Companies rely on a drop-off funnel, and an agent with infinite patience breaks it - thats the story.
For what it's worth, every one of these was a company not honoring something I'd already paid for, and I've filed exactly one arbitration. And support teams have LLMs on their side now too, so I'm curious where this goes once the regulators and arbitration queues get flooded.
Of course, it's impossible to know for sure what was LLM processed or not, but some of your posts (like this one) have been getting classified that way.
Got it. For transparency, all of my writing is my own but I do use LLM editing to improve the clarity of it and read every output carefully (and fully) before publishing.
On my own it often takes me 5-10 minutes just to figure out how to phrase something simple clearly. Like for example I don't how how to say that the amount of my own words I put into the LLM exceed (sometimes double or triple) the output that I end up posting online.
I'm afraid you're getting a bit piled on in the replies because readers feel strongly about this. But I want to add one thing re this:
> I do use LLM editing
What we're seeing is that LLM editing leaves nearly as many imprints on a text as outright generation, leading to two things: (1) it ends up sounding like AI, which most readers (at least on HN) don't want read, and (2) it ends up getting classified by the classifiers as genai anyhow.
What we're telling people (especially non-native English speakers) is to do their own writing and editing in their own voice and not to worry too much about minor mistakes - they are rapidly becoming signs of authenticity anyhow! Just worry about major mistakes, i.e. whether your main points are true.
While I appreciate your enthusiasm (truly), after two decades of writing I think I need to consider diminishing returns. I'd rather spend the time on thinking about what to say rather than how to phrase it.
how do I know which part is made up by the LLM or it's own interpretation or it's own opinion etc?
how can there be any value if I have to scrutinize every single word and decide whether it's truly what you intended to write or if it's mangling things, or if it's all just made up for engagement?
I've made this exact argument to other people, so I fully understand what you're saying here and agree with it.
I use the model in the same way a non-native writer would, for clarity. So it's critically important to me that there is no model interpretation or opinion added to my own.
I agree that whether it still counts as "my own" is up for debate. Made more complicated by the fact that I'm mostly using voice-to-text these days, so I'm dictating more than writing.
This website is a total garbage heap of haughty nobodies putting the brakes on actual conversation to make sure people are “following the rules” — as they break them with impunity.
Getting the OP’s comment flagged on some personal hunch is such a joke. For some reason this hive of nobodies is allowed to run rampant here, with no moderation on their own off topic comments and weird passive-aggressive conversation-ruining behavior.
I'm not going to pile on by posting a moderation warning in this context, but I do want to respond to one point you made:
> this hive of nobodies is allowed to run rampant here, with no moderation
You're quite right that there are many bad comments that don't get moderated on HN. The simple reason is that we don't see everything that gets posted here, or even 10%. There's far too much! It is humanly impossible to read it all (and would be, even if we didn't have many other time-consuming responsibilities, which we do).
Moderation is inconsistent, not primarily because we apply the principles badly (though perhaps we do—we're certainly imperfect), but because you're looking at a rather random sample of comments that have had moderator attention, and a whole lot more that haven't. Cnsistency is impossible (https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...), for much the same reason that not everyone who is speeding on a highway gets pulled over.
Therefore, if you see a post that ought to have been moderated but hasn't been, the likeliest explanation is that we didn't see it (https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...). You can help by flagging it or emailing us at hn@ycombinator.com.
Unfortunately, this all boils down is a bad look. Anyone who reads HN a lot will see a lot of bad posts, including many that break the site guidelines and get away with it. That is going to create annoyance and irritation, and a lot of that is going to get directed at the moderators for being a bunch of... well, I'll let you fill in your own pejoratives. Too bad for us, but it does go with the job.
Well your team never fails to flag my comments when I am espousing a viewpoint that goes against an established community view, such as manic, unhinged hatred of America’s current leadership, or if I’m critical of Europe’s ability to ship a meaningful volume software products.
I have no idea if you’re an ideologue or not, but my perception of the moderation team here is that political ideology and general SV narratives dominate your moderation strategy.
While we’re on the topic — it would not be very costly for you to employ a local LLM cluster and work queue to auto-moderate geopolitical topics that aren’t “tech enough” from ever seeing the light of day. Because the way it’s working now is the “I hate Trump” people spew the most insane flamebait imaginable, but they ensure ALL replies that disagree are flagged.
Clear thinking and rationality is not being prioritized here like it used to be. If your team is trying to do that, then whatever methodology you’re using is not working.
Its good that you're listening to the lefties, but are you listening to the right too? You know the guys I'm talking about. Something tells me you might be giving your own a pass on this very behavior you're accusing us lefties of.
Ok I'll bite, how do you reuse the answer to "what color is the sky?" while also layering in memory, custom system instructions and custom response styles?
It's probably only memory that'd need an answer for some form of caching to be worthwhile (though I'd be curious what that answer is) since memory is on by default and rarely identical between users.
The rest are nice to haves, but not everyone customizes settings just because they are there so there will be some large pool of users with the defaults who could hit cache without them.
I mostly agree but at the same time the change in underlying value from opus4.6 to Fable has not been as dramatic, and I'm not really sure if a "better Fable" is something most people even need. To the point where a faster and more cost effective model is preferable
Many would argue that opus5 is a regression in value despite what benchmarks say.
FWIW Seeing the same thing with Opus 5. Not sure if the models are being over optimized for agentic tasks or over reliance on synthetic data, because since Opus 4.6 I feel like emotional/conversational intelligence has been on a decline (Fable being the exception)
reply