HNHacker News
TopNewBestAskShowJobs

garethsprice

1,223 karma · joined July 7, 2011

submissionscomments
garethsprice··on The tipping point: what happens when deaths outnumber births?
Had a peek at Morland's book on Amazon and the headlines he pulls at the top of Chapter 1 for the impact of falling fertility on society:

'Russia running out of "single-use" soldiers; 'UK running low on fuel, truck drivers'; 'China's factories are wrestling with labor shortages'.

Are we supposed to feel bad that people aren't rushing to birth the next generation of cannon fodder, truck drivers, or work in factories for 60-70 hours a week? Are these roles that Morland would have his own children fill, or just other people's?

Genuinely bizarre choice of headlines to present a pro-natal argument with. Did he not read the actual book? Bad editor? AI slop?

garethsprice··on Googlebook
Another perspective from the posters saying "rich people"; in most advertising, it is aimed at aspiration and not reality - so "people who want to be rich leisure class people, or social media influencers", which tracks for a low-end laptop for a younger phone-native audience.

Advertising aimed at the actually rich is usually more about saving time, "elevated" experiences, or building legacy.

garethsprice··on Open Design: Use Your Coding Agent as a Design Engine
One use of design is signaling, but not all - successful design is that which fulfills its purpose.

Many designed things do not need to be differentiated and will benefit from a homogenous AI-powered design (internal documentation, local service business communications, etc) in the same way that desktop publishing replaced hand or type-written notes but did not replace professional designers (although it did require them to learn digital tooling).

For designs that do benefit from being differentiated it'll be interesting to see what happens. If anything, AI homogeneity provides more opportunity for talented human designers who can provide "design alpha" beyond whatever trends the LLMs sucked up in their last crawl.

garethsprice··on Opus 4.7 knows the real Kelsey
Failed for me - no identification of me by pasting text, and refused to search the web as it said that’d be a privacy violation. I have some writing around the Internet but not much, and less tagged with my real name. My guess is it limits itself to “public figures” defined as people who have a lot of publicly posted text.

I am glad to see I am not considered a public figure and aim to keep it that way.

I also had to go oddly far back to find a piece of long-form writing I had done that was truly mine and not tainted by an LLM edit pass which was a slightly disturbing realization.

garethsprice··on Facebook Has a Health Scam Problem
More like, society has a Facebook health scam problem (or just a "Facebook problem").

Facebook I'm sure does not think they have a problem, as they make a ton of money from running these garbage ads with no consequences.

garethsprice··on Running local LLMs offline on a ten-hour flight
This is what I am doing - it is rare that I'm in a situation with no Internet while traveling, but very often there is an intermittent connection. Using local models or even hosted foundation models is frustrating exercise in cancelled jobs and timeouts, but Tailscale + mosh + tmux is a really nice way to connect to a workstation and resume from where the session left off - or leave it running doing its thing and come back to it later.

Same with running my local dev environment's docker containers, now they run on that workstation and my battery life is far higher, treating my portable device as a dumb terminal.

garethsprice··on Running local LLMs offline on a ten-hour flight
Currently working on an Amtrak with XReal One Pro glasses and a ThinkPad bluetooth keyboard from my Macbook Pro that is folded up in the seat pocket.

They are "OK enough" that it will be a matter of taste if they are acceptable or not at this point for you to use.

For coding they work fine for me, terminal tools work particularly well as I can bump the font size up. IDEs and web browsing aren't bad either, it's about the equivalent of a single 1080p screen. They are nicer than hunching over a laptop for travel use but I still prefer a proper monitor when available.

The optics are a generation or two from being where they need to be to market these as productivity devices, but if you like being an early adopter with all the quirks that come with it, they're fun.

garethsprice··on LLM pricing has never made sense
> I don’t think it’s crazy to believe that people will also be running local inference on their phones in the next 5 years.

How about now? https://apfel.franzai.com/ (iOS/MacOS, runs the 3B param model already bundled for Siri) https://github.com/alichherawalla/off-grid-mobile-ai (Android, runs ~7B models on flagship phones at 15-30tok/s)

Foundation model investment feels like the bubble around fiber optics circa 2001 - a great technology being pushed forward by a speculative mania as it seems like it'll be useful in some profitable way, but nobody's quite sure how.

garethsprice··on Everything we like is a psyop?
This isn't new, Theodor Adorno was making the point in the 1940s that taste isn't something formed in a vacuum and then expressed in the market, it's something the market produces in you and sells back. Chaotic Good is just an extension of the same tricks from back when Eddie Bernays was staging fake suffragette marches to sell cigarettes to women in 1929.

The more disturbing implication is that it's not just "everything on the internet" is fake so much as the whole surrounding culture of consumerist societies - and that possibly every thought you have ever had and will have is a product of these influences.

Mark Fisher is a good read on whether we're even capable of comprehending what it would mean to live outside such a system at this point.

garethsprice··on Everything we like is a psyop?
The data around media literacy is terrifying.

"Fewer than 2 in 10 teens can correctly distinguish between news, advertisement, opinion and entertainment" https://newslit.org/news-and-research/news-literacy-in-ameri...

"82 percent of middle school students couldn't tell the difference between sponsored articles and real news stories.

Out of the Stanford college students that were tested, more than 80 percent couldn't identify biased content from independent news sources supported by groups like lobbying firms as being less reliable than a mainstream news source." https://universe.byu.edu/2018/02/09/studies-show-lack-media-...

garethsprice··on Picasso’s Guernica (Gigapixel)
It was a good question. Investigating why people like $popular_thing that you just don't get is a good way to broaden the mind.

I hope it leads to interesting discoveries and more art to appreciate for you!

garethsprice··on Allbirds announces pivot from shoes to AI, stock explodes 175%
It is not my field either except for amateur interest. My understanding is that because they are publicly disclosing it to investors, it's not fraud in the legal sense. Fraud requires deception. It could be that they do genuinely believe that their expertise in shoe selling translates to expertise in AI infra.

If you agree (or think that enough people will agree to increase the value of the stock), you can buy the shares. If you don't, you aren't forced to buy it - and as a penny stock with a market cap of $21MM, nobody's index fund or pension is going to be forced to buy it either.

The stock market sort of _is_ just karma farming - building belief in your company vision so that investors bring capital. Nearly all companies trade at a positive "P/E ratio" - that is, the price of a share is on average 10-20X higher than their current earnings - the difference representing pure belief that the company will earn more in the future.

The regulatory guardrails around it are not perfect, but striking the balance between allowing the public to participate in the growth of successful companies in a dynamic economy, and protecting them from scams, is a fine line. Being able to invest in publicly traded companies has made many middle class and working class people wealthy, the regulations do work for most people to keep them out of trouble while allowing them to participate. The people who get hurt are the ones who trade penny stocks like lottery tickets, which is less a regulatory failure and more the impulsive behavior that drives any kind of gambling.

garethsprice··on Allbirds announces pivot from shoes to AI, stock explodes 175%
This would be my bet - filing for an IPO or even direct listing is months of SEC review, audits, underwriters, due diligence, etc. A shell with an existing ticker gets you access to that sweet retail investor cash almost immediately, and with less of those pesky regulators asking if you actually have a functioning product or company.

This is more akin to a reverse merger than a SPAC (eg. Berkshire Hathaway being a failing textile mill and WPP being a wire basket company) except it is unusual to see it happen within an existing leadership team. They sold off all their IP so figured they might as well use the shell to try and cash in on AI hype, I guess.

garethsprice··on Picasso’s Guernica (Gigapixel)
I didn't either until I saw some of his early paintings and learned he was a brilliant classical painter from a very young age (eg. "Portrait of aunt Pepa", at 14). The later surrealism and abstraction weren't due to a lack of technical skill, but his deliberate choice to create a new visual language that would reflect the rapidly changing world around him.

That shift also reflected the era he lived in - one where visual arts played a central role in the cultural conversation - making him a true part of the zeitgeist that is hard to imagine now when visual art feels less central and more inward-looking.

A lot of what feels cliche now started with him, it only feels commonplace now as his influence was so massive.

Imagine being born in 1850 when everyone got around on horseback and paintings were realistic portrayals of people, landscapes, religious figures in muted tones. Impressionism (Van Gogh etc) arrives and is considered radical, then in 1907 you see _Les Demoiselles d'Avignon_ with its bright colors and abstract depiction of cavorting prostitutes. It would certainly provoke a reaction. The 20th Century had arrived.

garethsprice··on Author of "Careless People" banned from saying anything negative about Meta
Ordered a hard copy of the book, don't trust that an eBook version won't get revoked or edited at some point.

Timely given I just tried to sign into Meta for the first time in a year or two as I am being required to work on a Marketing API integration, got prompted for a video selfie(!) and my account is now in "Community Review" as maybe my expression was too grumpy about being required to present myself for inspection. Abhorrent company.

garethsprice··on I'm betting on ATProto
How does ATProto solve the problems that the last 10–20 years have shown seem intrinsic to all social media once it hits a certain scale?

For example:

A simple simulation of social networks rapidly reproduced three well-documented dysfunctions: partisan echo chambers, concentrated influence among a small elite, and amplification of polarized voices - creating a "social media prism" that distorts political discourse. Notably, all attempts at conscious intervention failed to help or made things worse. [1]

Rather than fostering closer relationships, the algorithms and structures underlying social media platforms inadvertently contribute to profound psychological harm - particularly among teenagers, who are disproportionately affected by curated online personas, peer pressure to present a perfect digital image, and constant notification bombardment. [2]

And from Meta's own internal UX research, surfaced in recent harm-related court filings: researchers described Instagram as functionally a drug, users as binging to the point of reward deficit, and the platform's role as that of a pusher. [3]

I've gradually opted out of social media over the last few years. That Meta internal research was the thing that finally pushed me to delete IG, the last social app I was still using. My life has been noticeably calmer and better adjusted since - which makes me skeptical that a better protocol, rather than a fundamentally different relationship with technology and socialization, is our way out of the current mess.

[1] https://arxiv.org/html/2508.03385v1 [2] https://scholar.dsu.edu/cgi/viewcontent.cgi?article=1222&con... [3] https://www.lieffcabraser.com/pdf/2025-11-21-Brief-dckt-2480... (p. 33)

garethsprice··on Police used AI facial recognition to wrongly arrest TN woman for crimes in ND
The vendor they used, Clearview AI, does not allow you to request data deletion unless you live in one of the half-dozen states that legally mandate it.

https://www.clearview.ai/privacy-and-requests

I have suddenly becomes very interested in New York's S1422 Biometric Privacy Act.

garethsprice··on Ask HN: Is Claude down again?
Working for me (Desktop + Code) but had a bunch of oauth issues this morning and it's been up and down like a drunken monkey on a trampoline all week, especially around start of biz times in the US.

The Pentagon spat is having a hell of a Streisand Effect on them; DAUs +180% (source: SimilarWeb), paid subs doubling. If past experience is any indicator, it'll settle down as they scale up (or they fall out of the news cycle). Google Trends search interest shows the story: https://trends.google.com/explore?q=claude%2Canthropic&date=...

garethsprice··on Workers who love ‘synergizing paradigms’ might be bad at their jobs
The headline says these workers "might be bad at their jobs," but considered in the context of Graeber's "Bullshit Jobs" thesis - that a huge chunk of white-collar work is pointless make-work for surplus labor - then in a hierarchy that rewards BS-fluency (which Littrell speculates), they are actually _good_ at their jobs.

The study measures analytic thinking as a proxy for performance, but that is only the right metric if the organization rewards individuals on the basis of their ability to make good decisions. Which anyone who has spent time in a corporate setting will know is often far from the route to success in such a setting, regardless of what the organization would say.

If your role has no concrete output and your organization rewards BS-fluency, you need a jargon that performs productivity without being too specific - so this argot isn't useless, it maintains a hierarchy that the BS-fluent can be promoted through. Not so much a rising tide but a blocked toilet backing up through the org chart. And BS-receptive workers are more satisfied with their jobs, because by their organization's actual values (versus whatever might be written in the mission statement), they're succeeding.

The BS-intolerant and analytically competent are less satisfied because they're the ones running into the blockers that the BS is covering for - or working through them only to discover that there's no tangible work to do under all the jargon.

The takeaway for me is: if you're interviewing somewhere and the hiring manager starts talking about "actualizing synergistic paradigms" instead of telling you concretely what the team shipped last quarter, it is likely one of those organizations. Places that can tell you plainly what they do are the places where your work will matter.

garethsprice··on The Brand Age
Patek and Rolex have generally always held value well. The internet has brought price transparency and created a secondary market that has made this true for most well-known brands these days, with the caveat that there is not a high degree of liquidity (yes you can sell a Rolex same-day at a pawn shop but you'll take a big hit, otherwise you are waiting a while for a buyer and navigating various risks like scammers).

So you likely won't get rich day-trading (or even collecting) watches, but besides the initial capital investment it is not as irrational as it seems at first.

Of course, that only adds to the signaling power for the classes where merely having money isn't enough, and being a "savvy investor" is a status marker (ie. the upper-middle and nouveau riche).

The Patek Nautilus PG mentions is particularly interesting as a status symbol as it has not only held value but, for various reasons, been an outlier that has appreciated drastically in value. It'd be gauche even on Sand Hill Rd to wear a hat that says "My investments beat the S&P 500" — but for the last few years, flashing a Nautilus (~413% investment return from 2016-2026 vs ~322% in the S&P) is the equivalent of that, a self-reinforcing cycle boosting its value as a Veblen good where the high price is the point.

garethsprice··on The Brand Age
Yes - though this isn't my theory so much as settled science. There are no substantial counter-theories to the concept that status desire is a fundamental and universal human motive.

As for falsification: you'd need evidence like subjects showing no well-being drop (self-esteem, cortisol) from lab-induced status demotions, or entire cultures genuinely indifferent to respect. Neither holds up in the data: https://pubmed.ncbi.nlm.nih.gov/25774679/

John Adams put it well in Discourses on Davila (1790): "The desire of the esteem of others is as real a want of nature as hunger — and the neglect and contempt of the world as severe a pain as the gout or stone."

garethsprice··on The Brand Age
It being an "acquired taste" is part of the appeal. A lot of high-end stuff is ass-ugly on purpose. If everyone liked it because it simply looked nice, you couldn't tell who's "in the club" of other rich people. Brands will attach elaborate stories and histories to objects to make people feel cultured that they have invested time in acquiring the knowledge, but really it comes down to in-group object recognition.
garethsprice··on The Brand Age
Humans are status-seeking creatures, and status is expressed through signaling. If you're rich and so are the people around you, money alone ceases to be a differentiator. Ultra-luxury brands appeal to this by adding hoops that money alone can't clear: time, loyalty, relationships. The signal shifts from "I can afford this" to "I was invited to spend my money here."

Lines outside Louis Vuitton are more down-market, aspirational luxury - an ultra-wealthy person wouldn't be caught dead queuing on a sidewalk. Patek and Ferrari operate at the level above, where the signal isn't wealth but access. (HBS calls Ferrari's version "deprivation marketing.")

Is it a game worth bothering with? Enough people think so to sustain billion-dollar brands.

(Of course, PG writing an essay about being too smart for fancy watches - while knowing a lot about them - is its own signaling game, just aimed at a different audience)

garethsprice··on Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU
Something like this? (Llama 3.1-8B etched into custom silicon delivering 16,000 tok/s, doesn't use much PCIe bandwidth):

- https://taalas.com/the-path-to-ubiquitous-ai/ - https://chatjimmy.ai/

garethsprice··on Hard problems in social media archiving
Would it make sense to archive every word every person ever speaks? At what point does archiving everything people do constrain their ability to live freely in the present?

Despite being in written form (decreasingly so), social media feels more like a private conversation in a public space - and like all such conversations, it deserves the right to decay, so that we do not all become prisoners of the dumbest thing we ever said.

The transformative work of curation - choosing which pieces to save, to turn into books, diary entries, or blog posts that record context for posterity - is a valid part of how archivists build the corpus of history. Harvesting all the raw data simply because we can is a dangerous road.

garethsprice··on Language a Wood for Thought: Susan Howe's Work
> Orpheus pounds on the soundproof glass. He won’t get through—he’s mythical.

Our industry would be improved by opening the door for Orpheus.

There's a lot of interesting thinking in the last 100 years of literary theory for this new age of stochastic parrots, good to encourage people working in tech fields to dig into it for newly-relevant ideas.

Howe appears to be an adherent of the Burroughs/Gysin cut-up (later fold-in) technique of rearranging language to reveal how meaning is made which I'm sure has some teachings for tech practitioners in there.

"the writing machine is for everybody / do it yourself until the machine comes" (Gysin 1960)

garethsprice··on "Hate brings views": Confessions of a London fake news TikToker
"While the social media revolution has come with extraordinary benefits..." curious what Khan would say those benefits are.

It's certainly something a lot of people find entertaining but I would not say there have been "extraordinary benefits" to society or individuals from the average UK adult spending 1h 37m a day[1] of their lives on social media platforms.

"While smoking cigarettes has come with extraordinary benefits, we've seen a surge in people misusing these products and getting sick due to a lack of guardrails, tobacco companies need to do much more to make their product healthy and discourage bad faith actors from developing cancer..."/s

[1] https://wearesocial.com/uk/blog/2025/02/digital-2025/

garethsprice··on Tell HN: I am afraid AI will take my job at some point
Good paper from 2015: "Why Are There Still So Many Jobs? The History and Future of Workplace Automation": https://economics.mit.edu/sites/default/files/inline-files/W...

tl;dr: Technology eliminates tasks not jobs, ie. automating routine work while making the remaining human parts more valuable: judgement, problem-solving, knowing what to build and why, communicating with a bunch of stakeholders.

This judgement and communication layer has been stubbornly hard to automate across every previous wave of tech and so it will be with this one.

Even if AI is capable of good judgement in a problem space, are the users of the AI able to ask the right questions to get it to express that judgement? (speaking from experience: no).

Banging out syntactically correct code loses value but communication and end-to-end ownership gains value; translating vague C-suite wishlists into working systems, knowing when _not_ to automate or use AI, navigating organizational constraints, understanding the domain you are working in, the organization you are working in, and directing technology in their service.

As long as there are capital owners they will have a need to exchange that capital for skilled, professional judgement in high value tasks, just the nature of what tasks are considered high value will adapt over time, as it always has.

garethsprice··on Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop
It's fun for me. This is a good enough reason to do anything.

I learn a lot about how LLMs work and how to work with them.

I can also ask my dumbest questions to a local model and get a response faster, without burning tokens that count towards usage limits on the hosted services I use for actual work.

Definitely a hobby-category activity though, don't feel you're missing out on some big advantage (yet, anyway) unless you feel a great desire to set fire to thousands of dollars in exchange for spending your evenings untangling CUDA driver issues and wondering if that weird smell is your GPU melting. Some people are into that sort of thing, though.

garethsprice··on Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop
HP G9 Z2 Mini with a 20GB ADA 4000, 96GB RAM, 2TB SSD, Ubuntu. Would get a Macbook with a ton of RAM if I was buying today, a full form factor PC, the mini form factor looks nice but gets real hot and is hard to upgrade.

Tools: LM Studio for playing around with models, the ones I stabilize on for work go into ollama.

Models: Qwen3 Coder 30b is the one I come back to most for coding tasks. It is decent in isolation but not so much at the multi-step, context-heavy agentic work that the hosted frontier models are pushing forward. Which is understandable.

I've found the smaller models (the 7B Qwen coder models, gpt-oss-20B, gemma-7b) extremely useful given they respond so fast (~80t/s for gpt-oss-20B on the above hardware), making them faster to get to an answer than Googling or asking ChatGPT (and fast to see if they're failing to answer so I can move on to something else).

Use cases: Mostly small one-off questions (like 'what is the syntax for X SQL feature on Postgres', 'write a short python script that does Y') where the response comes back quicker than Google, ChatGPT, or even trying to remember it myself.

Doing some coding with Aider and a VS Code plugin (kinda clunky integration), but I quickly end up escalating anything hard to hosted frontier models (Anthropic, OpenAI via their clis or Cursor). I often hit usage limits on the hosted models so it's nice to have a way my dumbest questions don't burn tokens I want to reserve for real work.

Small LLM scripting tasks with dspy (simple categorization, CSV munging type tasks), sometimes larger RAG/agent type things with LangChain but it's a lot of overhead for personal scripts.

My company is building a software product that heavily utilizes LLMs so I often point my local dev environment at my local model (whatever's loaded, usually one of the 7B models), initially I did this not to incur costs but as prices have come down it's now more as it's less latency and I can test interface changes etc faster - especially as new thinking models can take a long time to respond.

It is also helpful to try and build LLM functions that work with small models as it means they run efficiently and portably on larger ones. One technical debt trap I have noticed with building for LLMs is that as large models get better you can get away with stuffing them with crap and still getting good results... up until you don't.

It's remarkable how fast things are moving in the local LLM world, right now the Qwen/gpt-oss models "feel" like gpt-3.5-turbo did a couple of years back which is remarkable given how groundbreaking (and expensive to train) 3.5 was and now you can get similar results on sub-$2k consumer hardware.

However, its very much still in the "tinkerer" phase where it's overall a net productivity loss (and massive financial loss) vs just paying $20/mo for a hosted frontier model.

← PreviousPage 2 of 14Next →