Hacker Newsnew | past | comments | ask | show | jobs | submit | whstl's commentslogin

I run a quite large website and there are a few patterns.

The usage is extremely quick, and follows easy-to-spot patterns. We noticed a spike in bounce rate.

They never come from Google, and the bad programmed ones just crawl several pages at a time, faster than a user could do.

Then there's the crazy spikes in visits from specific countries, pretty much scraping the entire content. Often from pools of IPs. In some cases had 30% unexplained (meaning: it wasn't viral or a marketing campaign) random sustained increases in traffic.

There's also the fact they don't interact with the complicated widgets, so zero XHR requests other than analytics pings.

They also don't cause spikes in Google Analytics, so I assume it's blocked, but they show up in logs and in the internal analytics.

It's not enough to DDOS the website at all, but it's a lot of noise in statistics that we gotta learn to filter.


I understand both sides.

Quality must come from engineering. If you’re depending on a product manager to ask you that you can improve the quality of the code, you already lost.

So it requires soft skills, proper framing and ability to iterate quickly on quality-related tasks without leaving junk and multiple-versions behind.

But I completely understand push back for “doing improvements developers want to do”: A lot of developers confuse quality with familiarity or even complexity/verbosity. So business people have a reason to be reluctant.

And as an engineering manager I also had to push back several times. The thing that makes money is not the place to learn new skills, for example.


I think there's an argument that it could be cheaper and better for morale to let employees upskill while working on the thing that makes money.

It really depends on how mature the developer is.

If they have the soft-skills to do it, then by all means.

If not, they need to upskill their soft skills before tackling anything big.


From my POV, the main thing that's really broken with interviewing right now is the filtering process, even before candidates do a take-home test.

In the last few years I was the main tech interviewer for a 300-employee fintech.

For a specific position, one recruiter got around 150 applicants, selected 5 great ones, who did take-home tests and mild-tech interviews. Offers were made to most.

For the same role/salary, but from another queue, a second recruiter got around 900 applicants, cherry-picked about around 70 of them. Out of those, only 40 completed the 1h take-home test. Only 20 delivered it, only 10 implemented the requirements. Of the 10, all were unable to answer even basic questions.

This was concurrently, so it wasn't "affected by AI".

I didn't changed my methods and in fact I didn't even got close to asking hardball questions to the second group.

The second recruiter didn't get their contract renewed and left.


It should but apps don't let us decide.

An intermediate seems to be trying to fix it.

Is it ideal? No. But it's the spammers who are to blame.


You know, I would love a feature that lets me mark push notifications as spam, and optionally send them to Apple. The last part is important for a variety of reasons, one of which is that notifications can be end to end encrypted.

Spam filter push notifications.

Ideally enough spam reports for Uber Eat’s constant marketing abuse and they lose APNs access for the Bundle ID associated with the spam reports. For example.


You’re right of course, but Apple won’t do it - they’re happily running a two-tier system where Uber, eBay, Doordash can force spam notifications on you with impunity. All my settings for marketing are off - eBay still sends me notifications about coupons (and additionally there’s no way to actually contact them to complain, of course). Doordash won’t let me get delivery notifications without marketing notifications.

Apple could fully enforce their policies and fix this in a heartbeat, but they won’t.


like I said, you decide by muting or removing the offending app.

Fine, but that’s was clearly not enough to stop the spam, nor it was enough to satisfy everyone.

There are some apps I can’t afford to mute or uninstall, such as phone, transportation, communication and work. I wish I could, but I currently can’t, I’m not privileged enough.

“Punishment by Apple” in this instance is somehow the only response anyone had to misbehaving companies.


Same for things like Uber.

I do want to know when a car is arriving.

I don't want messages asking if I'm hungry.


Hi whstl,

Are you hungry? Open your Uber Eats app now for 10% off.

/this message sent through PalantirFinder -- from marketing and coupons to ordnance, we deliver everything!


The same issue happens on desktop but it requires zooming a bit more than 110%, and is screen-size dependent.

I don't believe this is a context problem.

Claude Opus 4.7 has a very large context compared to itself, but IME it is the worst at following instructions, and completely disregards the (small) preferences prompt, even in the first or second message, even if the messages are just a few characters long.

IMO this is entirely a training problem.


Isn't a large context window still a problem though? At the upper bound, the more you put in the more each sentence washes out within that window?

I’m not talking about large amounts of text, I’m talking about a couple sentences back and forth.

It disregards things like “no follow up questions”.

Haiku, for example doesn’t.

This bias is a very human thing, actually now that I think about it. You just disregarded the “even if the messages are just a few characters long”. :)


haha! yes i read too fast but i did read it and i took "message is small" to mean the message you want followed within the large context, not the entire context is just a small message.

funny though it is a case in point: language is hard. and i get to hide behind being "preoccupied" . i wonder if llms have their own sense of preoccupation hmmm.


It's probably some internal conflict between following the original training and following user prompts.

Also reminds me of the gremlin issue with GPT. An (internal) prompt saying "don't say gremlins" wasn't enough.


Codex compaction is way better imo.

I've had many long-running sessions and it doesn't suffer the same retardation (the act of delaying, slowing down, or hindering progress) that Opus does.

The quality stays consistent and it actually seems to follow the instructions, todos, etc. even after multiple compactions.


if you look at claude code, it now says compaction is happening constantly, which is likely why

If compaction is throwing away crucial prompting instructions even when it's at a 1% of maximum token usage (like my example), then it's a software bug, not an LLM artifact.

Doesn't compaction invalidate token caching, btw?

I don't see how this has anything to do with my message, sorry.

Are you assuming that "using a LLM" automatically means "vibe coding"?

Is it not engineering anymore even if you micromanage and relegate the machine to a better typist, following patterns and doing research around?


A team I worked with once optimized a website, and cut about 400k of cruft, down to under 200k total.

I did shit that was career-damaging, such as deciding not using heavy libraries that a platform team was pushing me to do (thus affecting their OKRs). I had to deal with designers complaining that removing JS didn't allow the special animation they wanted. The works.

We did it, and we managed to even score a few positions in Google.

Right after that, someone from Marketing added a few dozen random "pixels" to Tag Manager "because that's how Marketing is done", forcing us to adopt a cookie banner and bringing the lighthouse score to half of what it was.

Notice: The pixel was collecting data but it wasn't being used by anyone. There weren't ever any marketing campaigns involving the website.

Eventually the person was fired by the CEO for not bringing results, and we managed to remove the pixels, but by them the damage was done.


The blame for this is 100% on Google.

They constantly reward websites that are on a hamster-wheel of chasing the latest SEO trends, while penalizing websites that have actual information for not jumping through the same hoops.

A company I know operated TWENTY THOUSAND blogspam blogs out of a single server/IP. Google knew all along that this was happening (the companies had strategic partnerships) and never did anything about it.

The last thing they did anything significant was when, Panda in 2011?

At this point it's clear they're a monopoly and only care about websites who cater to their whims + making money. Search be damned.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: