Hacker News Reader

f38zf5vdton Ask HN: What has been bothering you lately?

Many successful and failed revolutions happened without an economic crisis. For example, the American Revolution was led by an oligarchical class upset about taxation. The American Civil War is another example where a portion of states felt economically threatened and initiated conflict. The Glorious Revolution was almost completely ideological in nature, revolving around royalty and religion. It simply requires some powerful fraction of the country to believe they'd be better off under another ruling party, no one needs to starve for it to happen.

People who are experiencing unrest and idleness are easy to recruit -- and if you hang out with people ages of 18-30, you'll see there's no shortage of them.

f38zf5vdton Southern Poverty Law Center indicted for fraud, money laundering

Certainly checkbook journalism has had a past in the USA, but I don't recall anyone ever being charged with a serious crime over it.

f38zf5vdton Claude Skills are awesome, maybe a bigger deal than MCP

It's baffling to me. I was already making API calls and embedding context and various instructions precisely using backticks with "md". Is this really all this is? What am I missing? I don't even understand how this "feature" merits a press release from Anthropic, let alone a blog post extolling it.

f38zf5vdton Meta Leaks Part 1: Israel and Meta

dang knows, it's his handlers who don't care.

f38zf5vdton The Rise of Whatever

open blog

> But yes, thanks: I was once offered this challenge when faced with a Ren’Py problem, so I grit my teeth and posed my question to some LLM. It confidently listed several related formatting tags that would solve my problem. One teeny tiny issue: those tags did not and had never existed. Just about anything might be plausible! It can just generate Whatever! I cannot stress enough that this is worse than useless to me.

The probabilistic machine generated a probabilistic answer. Unable to figure out a use for the probabilistic machine in two tries, I threw it into the garbage.

Unfortunately, humans are also probabilistic machines. Despite speaking English for nearly a lifetime, errors are constantly produced by my finger-based output streams. So I'm okay talking to the machine that might be wrong in addition to the human that might be wrong.

> It feels like the same attitude that happened with Bitcoin, the same smug nose-wrinkling contempt. Bitcoin is the future. It’ll replace the dollar by 2020. You’re gonna be left behind. Enjoy being poor.

I mean, you were left behind. I was left behind. I am not enjoying being poor. Most of us were left behind. If we invested in Bitcoin like it was the future in 2011 we'd all be surfing around on yachts right now given the current valuation.

f38zf5vdton Analyzing a Critique of the AI 2027 Timeline Forecasts

I think the author is right about AI only accelerating to the next frontier when AI takes over AI research. If the timelines are correct and that happens in the next few years, the widely desired job of AI researcher may not even exist by then -- it'll all be a machine-based research feedback loop where humans only hinder the process.

Every other intellectual job will presumably be gone by then too. Maybe AI will be the second great equalizer, after death.

f38zf5vdton What Google Translate can tell us about vibecoding

I'm using o3 and Gemini Pro 2.5, paying for the high tier subscriptions. The complaints I get are from native speakers -- editors and end consumers. The LLMs tend to overfit to the English language, sometimes make up idioms that don't exist, use false friend words (especially verbs), directly translate English idioms, and so on. I've translated several book length texts now and I've seen it all.

f38zf5vdton What Google Translate can tell us about vibecoding

It's funny -- I independently implemented the same thing (without vibe coding) and found it doesn't actually work. When I ended up with was a game of telephone where errors were often introduced and propagated between the models.

The only thing that actually worked was knowing the target language and sitting down with multiple LLMs, going through the translation one sentence at a time with a translation memory tool wired in.

The LLMs are good, but they make lot of strange mistakes a human never would. Weird grammatical adherence to English structures, false friend mistakes that no one bilingual would make, and so on. Bizarrely many of these would not be caught between LLMs -- sometimes I would get _increasingly_ unnatural outputs instead of more natural outputs.

This is not just for English to Asian languages, even English to German or French... I shipped something to a German editor and he rewrote 50% of the lines.

LLMs are good editors and suggestors for alternatives, but I've found that if you can't actually read your target language to some degree, you're lost in the woods.

f38zf5vdton Marines being mobilized in response to LA protests

50 people raiding a Nike store in LA isn't even beyond local news lol... you can find hundreds of videos online of this happening on a regular basis for the past decade.

https://www.nbcnews.com/news/us-news/group-50-people-shoplif...

f38zf5vdton How the US defense secretary circumvents official DoD communications equipment

> That's possible I suppose, but do you have any evidence of that

Yes, in the chat where a reporter was accidentally present, many of the messages were set to be disappearing. I don't know why anyone would do that if not to avoid recordkeeping laws.

> The images of the text chain show that the messages were set to disappear in one week.

https://apnews.com/article/war-plans-hegseth-signal-chat-inv...

Further, Project 2025 suggests bypassing federal record keeping legislation by simply holding in-person meetings without record.

https://www.youtube.com/watch?v=xxe55mU4DA8

Oddly, the Project 2025 training videos that presumably the members of the executive cabinet have seen say _not_ to delete messages or set messages to auto-deleting _because_ that would be in violation of federal record keeping legislation.

f38zf5vdton DOGE is building a master database for immigration enforcement, sources say

Wired reported that his might be happening previously, but this is a separate report assembled by CNN. Please kindly do not report as dupe.

f38zf5vdton Hegseth had an unsecured internet line set up in his office to connect to Signal

It also doesn't make it true. As other people mentioned, there doesn't seem to be anything backing up the claim and others have contrary experience.

f38zf5vdton Hegseth had an unsecured internet line set up in his office to connect to Signal

[flagged]

f38zf5vdton US Government threatens Harvard with foreign student ban

I didn't mention Nazism, I said _fascist salute_. The fascist movement that was popular earlier in the 20th century and led to Nazism, but is not exclusively limited to it. The adoption of the "Roman" salute was from Italian fascism, which the Germans borrowed.

f38zf5vdton Whistleblower details how DOGE may have taken sensitive NLRB data

Personal enrichment? There's already an enormous amount of evidence here to indicate that DOGE is working on behalf of a foreign nation state. It is seeming more and more likely that members of the DOGE team are simply secret agents for a foreign military.

> Within minutes after DOGE accessed the NLRB's systems, someone with an IP address in Russia started trying to log in, according to Berulis' disclosure. The attempts were "near real-time," according to the disclosure. Those attempts were blocked, but they were especially alarming. Whoever was attempting to log in was using one of the newly created DOGE accounts — and the person had the correct username and password, according to Berulis.

f38zf5vdton Trump's attacks on universities get darker, with shadows reaching our shores

"Let's do all of this, but much worse, with more corruption, and more violently" is a weird response to perceived political grievances. Like how Leninist critiques of the Russian monarchy and subsequent revolution lead to a system at least as bad as that of the Tsar.

If you want to deconstruct the status quo, make sure that your outcomes will be actually be better for you. Even Stalin ended up as a victim of his regime at the end of days, his physician ending up arrested and being interrogated while his health declined.

f38zf5vdton I'm the Canadian who was detained by ICE for two weeks

There's always https://news.ycombinator.com/active

It's embarrassing that to get to the content that technofascists don't want you to see, you need to visit a secret URL.

f38zf5vdton DOGE Exposes Once-Secret Government Networks

Is there some way to see these dupes? I always search first but when searching this URL, it didn't come up.

Seems I can't delete this now either so...

f38zf5vdton DOGE staffer is trying to reroute FEMA funds

> There have been no AI drones in that war. Drones are reigning free because nobody established air superiority.

This hasn't been true since last year, when automated targeting systems began entering the battlefield. They still are not as widely deployed as personnel controlled drones, but over time it appears to be shifting towards battlefield autonomy. Nuclear weapons are a cudgel, automated human-free weapons are a scalpel.

https://www.atlanticcouncil.org/blogs/ukrainealert/missiles-...

f38zf5vdton DOGE staffer is trying to reroute FEMA funds

I see. My opinion is that nuclear weapons are more or less useless in a modern conflict, as has been observed in the Russia-Ukraine war with endless Russian sabre-rattling. The collateral damage from nuclear weapons makes their usage suicidal, and in the case of the US, states breaking away from the Federal government would also inherit a nuclear arsenal -- leading to a stalemate.

The Russia-Ukraine war has convinced me that artificial intelligence driven drone warfare has rendered nuclear warfare obsolete. The drones act as a force multiplier which can mass target the enemy with little collateral damage.

f38zf5vdton DOGE staffer is trying to reroute FEMA funds

It seems like there is a winner-takes-all opportunity here to me. Since many of the billionaires in the US have made the decision to align with Trump, if many of the states unify and can overthrow the government, they also stand to separate the billionaires from their wealth.

The Federal American government looks like it's playing a dangerous game -- or at least makes the assumption that individual states within it will be unable to mount a concerted response. With inflation and unemployment rising, and hundreds of thousands of federal workers out of jobs, it's possible state governments may become radicalized.

f38zf5vdton DOGE staffer is trying to reroute FEMA funds

I think maybe the user is implying that a civil war will break out. I started reading about the original American Civil War after recent events. The lead up to that was protracted. As long as some states begin assembling militias now and unify, it's probable they will be able to mount an attack against the Federal government as happened back then.

https://en.wikipedia.org/wiki/Timeline_of_events_leading_to_...

f38zf5vdton Ingesting PDFs and why Gemini 2.0 changes everything

It works okay out to roughly 20-40k tokens. Once the window gets larger than that, it degrades significantly. You can needle in the haystack out to that distance, but asking it for multiple things from the document leads to hallucinations for me.

Ironic, but GPT4o works better for me at longer contexts <128k than Gemini 2.0 flash. And out to 1m is just hopeless, even though you can do it.

f38zf5vdton Did Elon Musk Appear to Sieg Heil at Trump Inauguration?

There must have been a breakdown somewhere in our objective view of reality when doing fascist salutes at the inauguration of the largest tech-business bearing country in the world, by the wealthiest tech CEO in the world, are whittled down to "Celebrity troll moves". I don't know if HN is beyond saving, but it does feel like it tows the line when it comes to techno-fascism.

A discussion worth having, if we could.

f38zf5vdton Ask HN: Teams using AI – how do you prevent it from breaking your codebase?

I have dealt with some codebases that were purely assembled with ML and they tend towards being completely unmaintainable nightmares exhibiting all the worst tenets of things like object-oriented code design. Levels of inheritance 8 modules deep, global variables being passed around everywhere to escape them, the works.

For ML driven code development, I find it works best when I used it to make pure functions where I know exactly what I expect to go in and out of the function, and the LLM can simultaneously write the tests for it to ensure that it works. LLMs do not plan like humans, even when finetuned they seem to have difficulty being integrative with knowledge beyond pattern matching.

That being said, 90% of coding is pattern matching to something someone made already. And as long as I'm writing pure functions and providing suitably adequate context for what the model needs to produce, LLMs seem to work wonders. My rule of thumb is to spend 10-20 minutes specifying exactly what I need in the prompt, and then tuning that if I fail to get the expected result.

f38zf5vdton It's not just obesity. Drugs like Ozempic will change the world

The current narrative of Ozembic reminds me a lot of the 1930s narrative around amphetamine. [1] The mood booster wonder drug was used extensively in the military and civilian population, from soldier to politician. Indications included weight loss, depressed mood, alcoholism [2], and fatigue.

> Another early pharmacologic approach to alcoholism treatment used psychostimulants to create a feeling of well-being and obviate the euphoric effects of alcohol. Bloomberg (1939) administered amphetamine to 21 alcoholic patients, who became more alert and energetic and reported no desire to drink (see also Bowman and Jellinek, 1941). Reifenstein and Davidoff (1940) found that amphetamine was of benefit in cases of acute alcohol intoxication and recommended it to treat depression in institutionalized alcoholics (see also Bowman and Jellinek, 1941).

Of course, I'm interested, but I will let medical science do the damage first before I elect to dive in.

[1] https://ajph.aphapublications.org/doi/pdf/10.2105/AJPH.2007....

[2] https://pmc.ncbi.nlm.nih.gov/articles/PMC4453501/

f38zf5vdton Differential Transformer

There is a downside, every attention layer has to effectively compute attention twice (run scaled_dot_product_attention). As scaled_dot_product_attention is usually one of the most expensive operations in training and inference of a model, it seems like networks using this may be substantially slow and perhaps should considered against larger networks with more attention layers.

https://github.com/microsoft/unilm/blob/master/Diff-Transfor...

f38zf5vdton Llama 3.2: Revolutionizing edge AI and vision with open, customizable models

Refer to the blog post I linked. Molmo is ahead of Qwen2 72b.

f38zf5vdton Llama 3.2: Revolutionizing edge AI and vision with open, customizable models

1. Ignore the benchmarks. I've been A/Bing 11B today with Molmo 72B [1], which itself has an ELO neck-and-neck with GPT4o, and it's even. Because everyone in open source tends to train on validation benchmarks, you really can not trust them.

2. The method of tokenization/adapter is novel and uses many fewer tokens than all comparable CLIP/SigLIP-adapter models, making it _much_ faster. Attention is O(n^2) on memory/compute per sequence length.

[1] https://molmo.allenai.org/blog

f38zf5vdton Blood puddles, mold, tainted meat, bugs: Boar's Head inspections are horrifying

Most first world countries have long and vivid histories of food safety scandals. In the US in the 2000s, the PCA comes to mind.

https://en.wikipedia.org/wiki/Peanut_Corporation_of_America

The UK had mad cow disease, Italy continues to deal with diluted and falsified olive oil, the EU has had repeated horse meat scandals. Most recently Japan had a health food supplement scandal in which many people died.