HNHacker News
TopNewBestAskShowJobs

karagenit

859 karma · joined August 12, 2018

http://caleb.software
submissionscomments
karagenit··on Claude Code reads AGENTS.md only when telemetry is on [fixed]
Yeah, this doc is really interesting. It has some absolutely claude-fried lines like:

> A hook is Koa-style middleware: it reifies an endomorphic continuation

> Five forms, each on tool.call;

But on the other hand uses distinctly human phrases like “one may” and “etc./e.g.” that I rarely see LLMs use. Not sure what to think.

karagenit··on Salesforce Global Outage
The bug was from 2009, so probably not :)
karagenit··on Corgi kills short-lived website that ranked its female employees
Way off topic, but free climbers definitely do use carabiners. Free here means “free from aid” i.e. not using ladders or other tools common in aid climbing, but free climbing still includes safety devices to catch a fall (usually involving carabiners). You’re thinking of free soloing which also excludes safety devices. And yea, these terms are confusing, I guess naming things is hard!
karagenit··on Breaking the WAL
I’m sorry you’re feeling ganged up on, I thought it was a nice article and cool you were able to repro the bug with your tools.

Like some others have mentioned, one of my earliest thoughts was “how much of a hint was the LLM given about the bug?” I think if the prompt used was stated clearly/verbatim near the beginning of the article, that would probably dispel a good amount of the criticism.

karagenit··on Building a Rust Inference Engine That Matches Llama.cpp
I think your natural writing voice is nice and the project idea is cool. There’s nothing wrong with using LLM coding tools. However, I have to read Claude output all day for my job, and it’s kind of bad at writing. I’ll be damned if I spend any of my free time reading more of it. I would really enjoy reading an article on this project from you directly. And not everyone else uses LLMs for producing technical texts (at work, we’re expected to understand the LLM output and put the correct/important parts in our own words when writing documents).
karagenit··on Apple engineer says he was fired after refusing to send cust. device IDs to AT&T
The employee didn’t have a contract with Apple so it isn’t a contract dispute. These would be statutory causes of action (also civil, but different than contract law) based on the claims in the suit. The plaintiff alleges Apple broke the law, but specifically whistleblower retaliation and disability discrimination statutes, not consumer privacy laws (which is sort of a secondary concern to the whistleblower claim, since those protections can apply even if no law was actually broken).
karagenit··on Show HN: Bribes.fyi – Know before you go. New feature added
It’s a cool site and you don’t seem to be spamming. But your account is new and has only posted links to your site (4 times) which probably caught the automated spam detection system. I’d gently encourage you to engage with the community and share some things that aren’t your own work as well. From the guidelines:

> Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity.

karagenit··on OverpAId – Fire your CEO. Hire the future
Can I ask how your orgs measured productivity before and after remote work/RTO? I agree that employees would have a much easier time accepting RTO if they were given hard evidence it makes the business more productive, but it seems like almost all companies don’t do this and simply justify RTO “because we know best” which frustrates ICs.
karagenit··on China's open-weights AI strategy is winning
Yep, and following that with

> and Chinese models are poised to take the lead.

makes it sound like the second part is a continuation of the first quote from the same source, but actually the second link is just some random person’s substack post from almost a year ago.

karagenit··on Top researchers leave USA for the Netherlands (in Dutch)
Tangent, but I’m really curious what country you’re from that uses the endonym for Göteborg but then also spells the capital of Denmark like Kopenhagen?
karagenit··on The truth about being a manager
> You will encounter business decisions you think are terrible but you still have to sell to your team. You cannot vent your frustration to the people you lead.

I actually disagree with this point a lot, as an IC. My manager shares his honest opinions with us, and I respect him more for it. It seems like the rest of the team feels the same.

I’ve had managers try to sell <obviously bad thing> as something good for the team, and it sucks. It feels like being gaslit. I think honest, open communication is a much better way to run a team. We’re all adults and professionals too; we can handle the truth.

karagenit··on Did Claude increase bugs in rsync?
It seems like based on e.g. [1] the article originally made some stronger claims about “no difference in bugs” that have been corrected. I agree that now it seems fine, but those edits might be why it feels like some commenters read a different article than you.

[1]: https://news.ycombinator.com/item?id=48418186

karagenit··on Zerostack – A Unix-inspired coding agent written in pure Rust
Curious how you’re handling prompt caching, as I understand it most LLM providers essentially inject tool definitions in the system prompt, so changing tools dynamically breaks the cache. This has been a big annoyance for me in a separate project; I currently just implemented my own tool-ish system that defines schemas in user messages and instructs the LLM to return matching JSON, but it’s less reliable than using the native tool calling + structured outputs available in the API.
karagenit··on I found a vulnerability. they found a lawyer
True, but the article also says:

> That's it. No rate limiting. No account lockout.

To me, if he confirmed that there’s no rate limiting on the auth API, this implies a scripted approach checking at least tens (if not more) of accounts in rapid succession.

karagenit··on Show HN: My attempt to formalize "Buddhist" reality
I would highly recommend giving this excelled LessWrong post a read: https://www.lesswrong.com/posts/rarcxjGp47dcHftCP/your-llm-a...

It isn't a perfect fit, since the article talks a lot about the scientific method which doesn't apply super well to philosophy+math, but I think there are some strong parallels here.

karagenit··on LiteAI – OpenAI, Anthropic, and Google LLMs at a discount
Looks cool! Does it support prompt caching? And do you have any data showing how your latency compares to going directly to the model providers? I’m thinking about trying it out but those are my two big reservations.
karagenit··on 'Ich Bin Ein Berliner': JFK Got It Right–He Was No Jelly Doughnut
https://archive.ph/i0WRW
karagenit··on The ‘white-collar bloodbath’ is all part of the AI hype machine
What if the number of game critics just hasn’t increased, and since they can only play/review a fixed number of games each year due to time constraints, the number that they acclaim each year hasn’t grown? Not saying this is necessarily the case, just suggesting the possibility.
karagenit··on Scraping Zillow for the Little Guys
Yeah, I'd probably go that route if I wanted to scrape more than the handful of pages I needed for this project. I wonder if it would work on Zillow or not. Even my simple workflow of "click the next button, save the request in devtools, repeat" was suspicious enough to trigger a captcha-type "are you a bot?" challenge. Maybe it was just too many requests quickly like you mentioned, or maybe they're doing something more advanced like mouse movement tracking.
karagenit··on Baby is healed with first personalized gene-editing treatment
Do you have a citation for this? The only relevant study I saw on the LISTEN website was a preprint of a study showing data on self-reported post-vaccine symptoms, but didn’t really talk about causes or gene edits (Krumholz et al. 2023).
karagenit··on Yi-Coder: A Small but Mighty LLM for Code
Yep, been waiting for the same thing. Maybe at some point it’ll be possible to use a large multilingual model to translate the dataset into one programming language, then train a new smaller model on just that language?
karagenit··on Notes on Buttondown.com
Hah at least it’s two/three letters, my personal site is at .software and most people get really confused by an eight letter TLD.
karagenit··on U.S.-Saudi petrodollar pact ends after 50 years
In terms of total petroleum products (including crude, gasoline, and diesel) the US has become a net exporter in the last few years.

> In 2020, the United States became a net exporter of petroleum for the first time since at least 1949. In 2022, total petroleum exports were about 9.52 million barrels per day (b/d) and total petroleum imports were about 8.33 million b/d, making the United States an annual net total petroleum exporter for the third year in a row.

https://www.eia.gov/energyexplained/oil-and-petroleum-produc...

karagenit··on US Supreme Court backs Starbucks over fired pro-union workers
> For the foregoing reasons, we hold that district courts must apply the traditional four factors articulated in Winter when considering the Board’s requests for a preliminary injunction under §10(j). We therefore vacate the judgment of the Court of Appeals and remand the case for further proceedings consistent with this opinion.

Correct me if I’m wrong, but it sounds like this is just about a temporary injunction to give the employees their jobs back while the actual labor case gets decided? The lower court used the two part McKinney test to grant the injunction, and the Supreme Court said no, you have to use the four part Winter test like other district courts, go back and decide again.

karagenit··on I'm a Former Surgeon General and I Couldn't Believe My $10k Medical Bill
I have zero medical expertise, but were the two blood tests and an xray really necessary for what seems like a simple case of dehydration?

Anecdotally, I recently got a simple blood test to keep an eye on something that’s been borderline in the past. Unbeknownst to me, my doctor ordered nearly 30 things to be checked in the blood panel, at $50-$100 a piece. Luckily my insurance negotiated a much lower price and covered most of it, but I can’t imagine paying $2,000+ just to check e.g. your cholesterol level.

I think the problem is alignment of incentives: for a doctor, over-testing has little to no negative consequences, but under-testing could lead to guilt due to a patient dying, loss of reputation because something was missed, lawsuits, etc.

As a society, is this how we should be using our limited medical resources? What if we made fewer xray machines and instead spent more on something like cancer screening? How many net lives could we save?

karagenit··on Value of one of S.F.'s biggest buildings plunges by 80% after Uber, Block leave
Arguably because it’s a two sided marketplace (matching drivers to riders) the network effect is very strong, so to have any success you have to dump a ton of money into marketing to gain enough users to make the app viable.
karagenit··on Evidence of price-fixing in the oil industry?
According to the EIA domestic crude production is at an all time high: https://www.eia.gov/todayinenergy/detail.php?id=61523
karagenit··on More Agents Is All You Need: LLMs performance scales with the number of agents
And it’s equally annoying to see this comment about that comment about the paper every time. Recursion!
karagenit··on What are the "worst" spelling bee pangrams?
Very neat. Would it be more efficient to start with every known pangram (~35k) and calculate the score from there instead of scoring every possible set of 7 letters (~8B) and then filtering for pangrams?
karagenit··on Rabbit Inc. Faked Their Demo (Again)
I don’t know, it was a pretty simple demo (just audio recording, transcription, and summarizing) so would it really be worth faking? The only things the article mentions are the typo (which actually makes me think it’s more likely to be real than faked) and the audio mismatch (which he explains on twitter was simply because he clicked an earlier recording accidentally - makes sense, I would do a dry run before actually recording the demo too).
Page 1 of 3Next →