HNHacker News
TopNewBestAskShowJobs

ashdksnndck

752 karma · joined May 16, 2025

submissionscomments
ashdksnndck··on AI got the blame for the Iran school bombing. The truth is more worrying
So to clarify, your argument is it’s ok to target civilians with bombs as long as they are located in a nation that practices terror?
ashdksnndck··on AI got the blame for the Iran school bombing. The truth is more worrying
The only reason Iranian bombs aren’t hitting America is because their range isn’t long enough. Iran-commanded forces (located in Iran, Lebanon, Yemen) have been targeting civilians for many years.
ashdksnndck··on The Los Angeles Aqueduct Is Wild
Can you clarify what you mean by: “they refuse to repurpose a downstream dam”

California has insufficient water storage to meet demand, it’s not like we have huge dams lying around that we leave empty when there is water available to fill them.

You might be referring to Don Pedro dam - but we are already filling that up (modulo what we need to keep empty for flood control). SF has some contractual right they could possibly exercise to water in Don Pedro but that doesn’t magically result in California’s water supply being held constant if we stop storing water in the Hetch Hetchy. If SF gets the Don Pedro water, that means someone else that was going to get it is deprived.

Now, you could argue that the state can get by with lower storage because ag needs to consume less or more groundwater recharge or whatever, but that’s a different question.

ashdksnndck··on The Los Angeles Aqueduct Is Wild
The country club is also negligible compared to agriculture. Farmers are politically not an attractive punching bag though.
ashdksnndck··on Wired headphone sales are exploding
> - They don't need charging. Charging may seem like a minor inconvenience, and we're used to charging a lot of devices. However, even a minor inconvenience is still an inconvenience.

I’ve always found it more convenient to keep the headphones plugged in when I’m not using them then when I am. The cord is not going to get tangled when it’s not attached to my body, and aren’t going to unplug themselves sitting on a shelf. Different strokes though.

ashdksnndck··on 1M context is now generally available for Opus 4.6 and Sonnet 4.6
My change cuts across multiple systems with many tests/static analysis/AI code reviews happening in CI. The agent keeps pushing new versions and waits for results until all of them come up clean, taking several iterations.
ashdksnndck··on The MacBook Neo
I think for this plan to work you’d have to force the developers of Xcode to work on the 4 Gb machines first. If they could do that, the rest would follow naturally.
ashdksnndck··on I'm glad the Anthropic fight is happening now
I always listened to the podcast and forget they even have videos. Have a hard time imagining myself sitting and watching a 2 hour interview when I could listen while exercising or doing chores. Am I missing anything?
ashdksnndck··on Meta acquires Moltbook
I’ve been wondering if ChatGPT is actually coming up with the idea of posting to Reddit when the user is asking a question and ChatGPT can’t find a good source to answer it. ChatGPT has never suggested this to me, but it wouldn’t be a completely crazy thing to do. A lot of ChatGPT answers are sourced from Reddit (via search, and also via training data). If everyone starts asking ChatGPT everything instead of Reddit, there won’t be as many new conversations happening. Promoting users to post questions to Reddit would help solve the user’s direct problem, and also make the ensuing answers available to ChatGPT to help with future conversations.

I understand that a lot of people would be very unhappy if this is true, but I can imagine from the perspective of a product person at OpenAI that it helps them in multiple ways.

ashdksnndck··on Don't post generated/AI-edited comments. HN is for conversation between humans
I don’t respond to specific comments with accusations, because I can’t prove it and it would suck to be falsely accused. But I find it really depressing to watch deep comment threads with someone debating with an AI. The human is putting so much effort in, and the AI is responding with all these well-written but often flawed arguments. I wish I could do something to save that person from that interaction.
ashdksnndck··on Meta acquires Moltbook
There are also tons of comments written by AI on hacker news. There are whole discussions between AI bots arguing over whether AI is a sham.
ashdksnndck··on LLMs work best when the user defines their acceptance criteria first
I imagine there would be value in not just throwing all of GitHub commits in as training data, but also rating the quality.
ashdksnndck··on LLMs work best when the user defines their acceptance criteria first
I think this is in the training data since they use commit data from repos, but I imagine code deletions are rarer than they should be in the real data as well.
ashdksnndck··on Where things stand with the Department of War
It’s not that big of a coincidence that the countries that superpowers want to conquer, and need defending, are neighbors to the superpowers.
ashdksnndck··on MacBook Neo
I wonder if this can develop iOS apps.

For me an iPad with the Magic Keyboard case is already my personal laptop, and this device would be a downgrade in almost every way (ergonomics and specs).

The one thing I can’t do with my iPad that interests me, that’s got me thinking of buying a personal Mac, is develop software for iOS.

ashdksnndck··on When does MCP make sense vs CLI?
Verbosity of the output seems orthogonal to the cli vs mcp distinction? When I made mcp tools and noticed a lot of tokens being used, I changed the default to output less and added options to expose different kinds of detailed info depending what the model wants. CLI can support similar behavior.
ashdksnndck··on Microgpt
It already is in some threads. Sometimes you get the bots writing back and forth really long diatribes at inhuman frequency. Sometimes even anti-LLM content!
ashdksnndck··on I am directing the Department of War to designate Anthropic a supply-chain risk
Ghengis Khan didn’t need your chest of gold, he owned many gold mines. Regardless, he was going to take it from you the easy way or the hard way.
ashdksnndck··on Banned in California
In many cases, California’s environmental regulations don’t make an earnest attempt to permit safe ways to do things.
ashdksnndck··on Banned in California
I assume you use semiconductors yourself, since you are posting here. But you want their manufacture to be banned in your state.

So the right thing is to outsource the dirty jobs to countries that can’t afford to be picky?

Wouldn’t it be better for the world if we used our wealth to develop methods of safe semiconductor manufacturing with low environmental impact, and proudly built those facilities in California?

ashdksnndck··on The EU moves to kill infinite scrolling
Netflix (even before they introduced ads) optimized for watch time. Higher watch time = higher retention for subscriptions (even when prices go up).
ashdksnndck··on OpenAI has deleted the word 'safely' from its mission
Is Google actually in front? I know Google keeps publishing impressive benchmarks but developers who are the most engaged and demanding users of LLMs keep choosing to use Claude instead. My uninformed take is Google is optimizing to the benchmark more vs. building a better product, which matches my overall impression of management at Google.
ashdksnndck··on Agent Skills
I had a kind of visceral distaste for all of this rules, skills etc stuff when I first heard about it for similar reasons. This generalized text model can speak base64 encoded Klingon but a readme.md isn’t good enough? However given the reality of limited context windows, current models can’t consider everything in a big repo all at once and keep coherent. Attaching some metadata to the information that tells the model when and how to consider it (and assisting the models with tooling written in code to provide the context at the right time) seems to make a big difference in practice.
ashdksnndck··on FBI couldn't get into WaPo reporter's iPhone because Lockdown Mode enabled
Doesn’t matter if the agent believes you. Only matters if the court jails you on a contempt charge.
ashdksnndck··on FBI couldn't get into WaPo reporter's iPhone because Lockdown Mode enabled
You think iPhones aren’t multi-user for technical reasons? You sure it’s not to sell more phones and iPads? Should we ask Tim “buy your mom an iPhone” Cook?
ashdksnndck··on Agent Skills
Is your view that this doesn’t work based on conjecture or direct experience? It’s my understanding Anthropic and OpenAI have optimized their products to use skills more efficiently and it seems obviously true when I add skills to my repo (even when the info I put there is already in existing documentation).
ashdksnndck··on Agent Skills
We’re working with the models that are available now, not theoretical future models with infinite context.

Claude is programmed to stop reading after it gets through the skill’s description. That means we don’t consume more tokens in the context until Claude decides it will be useful. This makes a big difference in practice. Working in a large repo, it’s an obvious step change between me needing to tell Claude to go read a particular readme that I know solves the problem vs Claude just knowing it exists because it already read the description.

Sure, if your project happened to already have a perfect index file with a one-sentence description of each other documentation file, that could serve as a similar purpose (if Claude knew about it). It’s worthwhile to spread knowledge about how effective this pattern is. Also, Claude is probably trained to handle this format specifically.

ashdksnndck··on Margin Call
Are you buying apps in the App Store or paying for digital goods (eg. signing up for streaming services) through iOS apps? Apple charges a 15-30% rent on all of that.

Apple also tracks and charges for conversions on ads for mobile games (eg. purchasing lootboxes on clash of clans) which makes them direct competitors with the other big tech ad platforms.

ashdksnndck··on First, make me care
My feeling about Gwern is that I won the jackpot if he happens to have written on a subject I want to know more about. His writing is a wealth of information. It not always compelling if I’m not already interested.
ashdksnndck··on Claude's new constitution
I’m not sure the inability to anticipate terror ending would improve the experience. Tricky one.
← PreviousPage 5 of 9Next →