Literary fiction author (‘Brat’ & ‘The Complete’). Used to be in tech, still coding for research (ie mucking about). Interested in both, where they meet, and what each can learn from the other. It’s all just language
twitter.com/gabriel666smith
henrygabrielsmith at gmail
投稿
Show HN: Audio Player with "Binaural Beats" tuned to the same key as your music
github.com
21 ポイント·投稿者 gabriel666smith··5 コメント
Show HN: The Port Augusta Times – "All the news that's fit to generate"
henrygabriels.github.io
1 ポイント·投稿者 gabriel666smith··0 コメント
Show HN: Create a clone of your old self from your old iOS backup; talk to it
github.com
3 ポイント·投稿者 gabriel666smith··0 コメント
Fmllm: 4mb training data, 100mb model, Fibonacci embeddings, near-coherent. WTF?
github.com
37 ポイント·投稿者 gabriel666smith··27 コメント
FMLLM: 4mb training data, 100mb model, Fibonacci embeddings, near-coherent. WTF?
github.com
2 ポイント·投稿者 gabriel666smith··5 コメント
Show HN: Scratch Radio Ft Doctor Von Peel: An LLM / Cat-Based Apple Music Client
This looks great for accessibility specifically. My fiancee is partially-sighted, so (while not directly related) I'm extremely supportive of developments in assistive tech. There's a lot of difference to people's lives that can be made.
For me personally - maybe because I'm not a very nice person, or maybe it's an association that many people would make - clever as the pun in the product name is, it does make me think of saying "mousepad" with a lisp. This seems like a non-ideal association.
>So the main reason for a windows view is to use Grok's subscription over the API correct?
Yeah, spot on. I literally wrote a Chrome extension to download -> grab last frame -> navigate -> reupload as new start frame, lol.
But - honestly - unclear how long video models will remain subsidised via subscriptions used in a browser or app, so I don't 100% know that it's a feature worth your time building.
>In the editor, you can grab "screenshots" of specific moments of the video you can then use as first frame for another AI video
Oh awesome! That's really great. I feel like such a moron lazily dragging the "screenshot" tool around Final Cut right now. Genuinely sick feature. Excited to use next time I'm doing stuff with video.
Nice! Because the Grok subscription is the cheapest-per-second video offering (that I'm aware of) but doesn't provide API access, I'm often moving between the browser and editor (typically Final Cut).
It feels cheaper by orders of magnitude right now per-second. I'm not sure how long that will continue, but until then, that's pretty important to me, especially when video generation is so imprecise, and thus more of a "rough draft" creative tool than a "final shot" tool.
Quite often my workflow looks like: "Grab this frame from which I'd like a continuation" -> "Upload back to Grok on web" -> "Generate until happy" -> "Download and use".
So a feature request (or I'll just fork next time I'm doing something with AI video) is simply a browser tab alongside the generation tab, literally just to make that window swapping easier, and ideally with 'download' pointing to a sensible location to avoid file replication.
This feels closest to how I use AI video, which is just kind of like an infinitely extensible library tab on my video editing software. I know you said 'no webview' - I don't know Swift well enough to know if that excludes this feature.
There's probably a bunch of ways to improve my specific flow once there's a browser within the UI, especially with MCP integration. Much to consider.
And overall it looks fun, and I'm excited to give it a try. It's great this has beat detection specifically, because that's non-negotiable in a video editor for me when doing AI stuff, as the medium remains more suited to non-dialogue work.
Similarly, the CapCut 'AI video editor' that's most similar to the 'MCP' feature is unusably chaotic, so I'm interested to see how well models do in your architecture. I can understand why you've found structures like 'beat' or 'transcription' massively aid it.
Final Cut export - if it isn't supported by what you've already done (I don't move stuff between editors much!) - is really useful to me personally also.
Best of luck! Excited to test it! Thanks for sharing.
Yes - I think the idea of 'citizenship' feels increasingly abstract and outdated to anyone who only rarely noticeably interacts with the entity they pay taxation to, typically solely in moments of extreme crisis. 'Noticeably' here meaning I'm excluding ambient aspects of life, like walking on a government-maintained pavement.
I would argue those 'non-essential' services - those which you actively notice benefiting from - are extremely important for building institutional trust in the state. For young people entering the workforce today, who (for example) are a demographic that have a lower requirement for the services of the NHS, it must be genuinely confusing and alienating to be paying (what I believe is) a historically high tax rate.
Citizenship, as an idea, does need to demonstrably benefit both the state and the individual. Otherwise, the state is basically asking people to draw on their reserves of empathy to have a reason not to act like a dick. That will, unfortunately, have much more varied results than empathy alongside tangible incentives.
The cost of not providing those 'non-essential' services may ultimately be larger, and more difficult to calculate, I think, than are calculable today. How you provide them without the cash to do so is a question I'm not smart enough to answer. But people have now - completely anecdotally - gone a very long time with low tangible incentives toward citizenship being provided by the either the state or local authorities, and the vibe is very, very bad.
I don't think this is restricted to the UK, either. The OECD’s 2024 trust survey found only 37% of respondents across 30 countries were confident their government balances the interests of current and future generations. [1]
The metadata matters in a different way. A single piece with a single composer can be interpreted very differently, over time, by different conductors and orchestras.
This means classical music is very badly organised in most streaming apps. Sorting by "artist" doesn't really work if the artist is listed as an orchestra, which has probably recorded the work of more than one composer.
Roon is another example of a streaming platform that treats this metadata with the same importance.
Lol, I'm personally too broke for a Spotify subscription - as in I couldn't sign up this month with the cash in my bank account - and that has often been true during the periods of my life I have not spent working in tech. I don't think I'd even be in their youngest adult user demo anymore!
We definitely exist. It's pretty tight out there right now. The £10.99/month I see quoted buys quite a few kilos of dry pasta.
In any case, the core argument I was making specifically is that I believe human curation can provide more friction for the audience than algorithmic curation tends to, and that this is a net-good for cultural education.
I would contend with classical music being the highest strata of musical culture (and the deeper idea of culture having 'levels'), but that's subjective. So, that aside, if classical music is your thing, and you believe:
> "The quality of podcasts and music streaming is on a quality level which radio never reached."
I'd recommend Glenn Gould's The Idea of North, which is one of my absolute favourite pieces of music. It's considered pretty innovative and important:
Glenn Gould wrote it specifically for radio, as it was commissioned by the CBC. So "government produced sewage". :-)
I'd also strongly recommend you give BBC Radio 3 (still often excellent, especially late at night) a try. NTS has some great classical shows, too. Perhaps you're already familiar with all of these, but - if not - they might give you a reason to turn on your radio more frequently than annually.
You're mostly correct - I was absolutely generalising, using internationally known examples (like the BBC), to be more comprehensible to the international audience here about cuts to public arts spending.
The BBC was a silly example of me to use (as was 'public radio'!) because you're right - it's mostly not funded by general taxation.
That said, the license fee is directly tied to inflation (via the CPI). Whether you see that as a correct measurement is another question. [1]
The BBC has also historically been funded partially by taxation. For example, taxation provided free television licenses to over-75s, until 2020, when that funding was cut, and the cost of doing this was passed to the BBC itself.[2]
But yes - quibbles aside, a sloppy example for me to have used there. Your sibling comment has some great sources which better illustrate changes to arts funding, compared to the off-handed ones I initially provided.
Yeah - I used "austerity" as a slightly sloppy shorthand in my comment. My two closest local libraries were closed due to cuts made during "the credit crunch" by the coalition government (from memory!), when "austerity" was the buzzword, so I felt it'd communicate most broadly (to HN's international userbase) the phenomenon I was gesturing at.
From what I can understand, these sources are saying: "because costs are up, those past cuts have not yet really been reversed, especially not in terms of real-terms funding per-capita, or funding for 'non-essential' services".
Super bleak reading - it's a very dark rabbit hole!
I work as an author. I believe this is total bullshit, from beginning to end - the ruling, the settlement, and the suit itself.
In the UK, we have a thing called the Public Lending Right [1]. This pays authors a fixed sum each time their book is taken out of a library, up to a capped amount.
The cap isn't very high - about $7k - so it is both an OK bit of income for authors who might be making very little money elsewhere, and also doesn't end up all going to authors who are already bestsellers. It's a decent legal system for helping libraries hold niche titles as well as the popular ones. This is, after all, the purpose of a library.
To establish my bias here: My debut novel came out after the period this specific suit concerns. I also uploaded it to LibGen myself.
I strongly believe that books should be available to read, free of charge, to all people. I benefited enormously from libraries and piracy growing up. I think they serve an important educational purpose that does not end when a person leaves school, and I do not think wealth or disposable income is a fair way to decide the breadth of a person's education.
I also have no problem with people making new "language things" using my work. I love sample-based music (like dance music, hip hop, etc) and it'd be hypocritical for me to take issue with anyone doing analogous things using books. Maximising sales is not the end-goal of making art, for me personally. Other artists feel otherwise. They consider training on pirated books stealing. That's OK - it's not for me to tell them what to believe.
The problem for me is that these corporations - undoubtedly still pretraining on pirated material - are, essentially, leeching. By not releasing the model as open-weight, freely available, they are not acting in the same spirit of the system they took advantage of. It's the Spotify model: pirate first, pay a nominal amount that does not meaningfully harm profit later. Now the dust has settled there, we can see the harm it has done to music culture.
A single settlement which does not establish precedent does not solve anything. A tokenistic $3k allows anti-AI authors to wave a cheque in the air and declare a victory. It pays the rent for a month or two. It does nothing for the months after that, when the corporation is still profiting. It does nothing to establish precedent for future artists, who also have to pay rent.
It would be (non-trivial, but) relatively simple to integrate - for example - download figures from Anna's Archive into the PLR. I'd happily dilute my PLR payment appropriately, because I think libraries are important.
You can't stop people pirating digitally replicable things. Digital ownership is not a concept that has held, or will hold.
There are only 23,000 authors in the UK who claim the cash from the PLR. To pay all those authors the national living wage in the UK (£26k) from the PLR, you would need to raise £546 million. That is around 1/34 of Anthropic's reported annual revenue.
I'm of course not arguing Anthropic should be solely responsible. But it's very frustrating that all the pieces of the puzzle for actually paying artists in a sustainable and ongoing way now exist, and one of the major obstacles to this - and the idea of a genuinely free, legal, international library, which creates more authors, writing better books, full-time - are legacy rights holders who remain attached to a completely dysfunctional and outdated concept of ownership.
So - unless part of a sustained and reasonable campaign, which understands the futility of (and damage to the medium and its creators caused by) treating digital ownership in the same way as physical ownership - this suit is close to pointless, and arguably actively harmful in the long term.
It is a shame. I think it’s especially a shame for children and young people.
A counter to radio remaining relevant is “I can listen to whatever I want, via this digital technology”.
Fine - but young people don’t yet have a complete mental map of what culture they enjoy.
In the UK, funding for libraries and the arts (public radio, the BBC) has been increasingly cut in the name of austerity. I think that’s true in the US too.
Despite this, young people have an unprecedented amount of access to culture. Or, at least, any form of culture that can be digitally replicated.
Soulseek and Anna’s Archive are wonderful projects. But they only have as much utility as the person using them can put into the search bar.
We’re increasingly entrusting cultural education - what to put into the search bar - to algorithms that are designed not (as the BBC’s mission statement is) to “inform, educate, and entertain”, but to retain attention.
Even worse: many of these (eg, Spotify) hide real access behind subscription fees. If you grew up broke, like I did, you will understand how inaccessible this real access is.
The most motivated young people will continue to find the workarounds piracy offers, and to find other routes for cultural education.
Any young person who isn’t in that upper percentile of motivatedness might increasingly be left out. You can attribute that to a choice they have made - I think that’s wrong.
I think the joy of broadening one’s horizons is available to all, and is also difficult to learn and a strain to continue to practice. I think it must be taught.
I think this is especially true and important when everyone has a constantly-available option of algorithmically-selected pleasantness.
Human radio DJs are not optimized for pleasantness, or to serve you relevant ads based on your conversations with them. At their best, DJs are educators.
Anyone who grew up listening to radio knows that, and knows the pleasure that comes from learning not to change the channel when something new and uncomfortable-feeling comes on, just in case it ends up being something you fall in love with.
You can personally work out how apply this one in engineering today!
If you have a spare hour or two, I'd encourage you to have a go at learning (however you learn best - I like just asking smarter people or robots stupid questions) what the maths means and why it's important.
And then once you feel like you have a vague grip on the principles, think about a problem in a domain you know a lot about. Try to see if the maths - and how it's changed our perception - could be used as a tool to solve that problem, or if the solution is analogous to a solution you could try in your own domain of expertise.
LLMs are good at speeding up, I think, the journey an idea has to take between "theoretical academic stuff for academics" and "a usable idea for regular people", because they increasingly allow you to ask an infinite number of stupid questions and give you (hopefully) reasonably good responses.
I've had loads of fun doing this today - specifically seeing if the idea this counter (from what I understand: a many-to-one conversion that kind of does and kind of does not preserve meaning) can tell me anything about the relationship between language and meaning.
I'm sure everything I've done today while mucking around has been the equivalent of a monkey with a typewriter (and Codex), but I think the huge breakthrough(s) you ask whether we're on the cusp of are relatively dependent on how many monkeys are throwing typewriters at problems they know a little bit about, after learning a bit about new ideas like this one. Historically, that's a really good way for broad cultural innovation to happen - distributed information applied across multiple domains by experts in them.
I don't think this is necessarily going to prove to be true.
I often see the sentiment: "the Chinese strategy only makes sense in the context of undercutting American labs' profit margins".
If, for example, you are a company with a near-monopoly on "serving video content", and you feel reasonably confident about retaining a decent slice of the serving-video-content market (Google in the west is an example, Tencent in the east), then training video models on your dataset - and releasing them freely - makes an awful lot of sense.
Free tools to create with mean more video content. In this hypothetical, you're reasonably certain that any video content which does get created will also be watched on your platform.
That is a net positive. The question becomes: How many watch-hours earns back the cost of training a model? It's probably not really that many, especially when you have a near-monopoly on a billion sets of eyes.
It's also a net-positive if people build better video models from research you release, because - again - you are reasonably certain that the even-more-innovative content those models produce will be watched on your platform.
It really begins to make strategic sense if your company is in a GPU-poor environment. Your costs cease at the point you upload a model if your users are running it themselves. You don't have to serve the model. The content is still created.
You are also less likely, I think, to alienate human creators whose work the model was trained on if the model is not sold back to them as a subscription, or by the token, but given for free as a tool.
This frames the conversation very differently. It creates, I think, less of an "us vs them" dynamic, and more of a rising tide.
It's true that it is also beneficial that these models undercut (especially in language models) American companies. But, generally, Americans are not the customers of Chinese companies releasing models. They are already serving a huge volume of customers in a complex, existing marketplace.
The full picture is much more nuanced than simply a geopolitical desire to undercut US labs, and there are several other reasons the strategy can make logical sense.
It's an interesting question, I think, because GGP is correct to say that geopolitical discussion dominates posts about work done by Chinese developers and organisations. That work is fascinating and innovative.
On the one hand, those orgs are tied to a global superpower. I couldn't name a single global superpower that has existed - ever - which hasn't committed acts I personally consider deeply immoral. I don't think we typically discuss OpenAI's work, and learn from it, in quite the same way we discuss the work of Chinese labs. OpenAI is a US defence contractor, and - whether you agree or not - it must be acknowledged that a lot of people would see actions like the invasions of Afghanistan or Iraq as deeply immoral. I don't think "which country is more evil?" or "which org is more embedded in its government?" is a productive route for this discussion - to me it seems like whataboutism. Others would disagree.
On the other hand, technical and artistic innovation has been sponsored and patronised by wealthy organisations (nation states and religions, predominantly) throughout history. That wealth was, generally speaking, acquired very violently. What does this do to the moral status of the work itself? I don't have an answer to that.
I do think I have a lot more in common with the average Chinese developer than I have in common with the average British (or American) political leader. I also believe the average Chinese developer has more in common with the average American developer than they do with the average Chinese political leader.
Lots of people might disagree with me about that, which is fine. What would be more difficult to disagree with, I think, is that I personally think I can learn a lot more from talking to people about the work we share than about the politics we do not.
Some examples: I personally think the Kimi line of models have been, for some time, by far the strongest of any lab's at creative writing tasks. I don't know how they've managed that. I'm completely obsessed with Bilibili ("China's YouTube"). I think many of its features are outstanding, and unmatched by anything a western video platform currently offers. The video content hosted there is often innovative, and occasionally artistically brilliant; the AI video usage specifically is notably more artistically advanced than anything I've seen on western platforms.
I'd love to listen and talk to some of the people who worked on this stuff, or who know more about it (perhaps in a less off-topic place!). I'd really love to work alongside those human beings. Because of geopolitics, that isn't easy. That is really frustrating to me. HN would, I think, in an ideal world, be a place that could happen.
I don't think you're taking the good-faith reading of the sentence's grammar here, which is:
"[When China is mentioned] the comments section stops discussing [technical aspects], and instead starts going on about [non-technical aspects], and all that [stuff which I do not feel is relevant]."
I'm struggling, though, to find a good-faith reading of what you meant by:
"Denying human rights. Classic. Sorry - you instantly lost any respect I could have maybe had for your opinion."
Specifically, "classic" - classic what? "Classic" meaning "a trait you personally believe is more prevalent in people who (might be) citizens of a specific country"?
Hopefully the good-faith interpretation of GP's comment, which you might have unintentionally missed, will help you engage with their opinion with more respect.
I think it's an easy opinion to empathise with personally. I would be enormously frustrated myself, on an individual level, if interesting technical achievements made by companies in my own country were overshadowed by political discussion about the actions of the country overall, when spoken about on technical forums.
This would be especially true for me if I - as an individual - disagreed with politically with my country's leadership. Frustration at a topic of discourse is not mutually exclusive with being in favour of human rights.
I don't think GP has done themselves any favours with the broad statements that close their comment, and is clearly frustrated themselves. That there's mutual frustration is probably a sign that empathy on both sides might lead to a really productive discussion - but focusing on the more substantive points they've made, rather than escalating by misinterpreting the less important aspects of what they commented - is what will lead to that.
Word. In terms of game level design, I think SSX Tricky’s Tokyo Megaplex is one of the most interesting, complex, simple, comprehensible, fun, beautiful, and downright cool pieces of work of all time.
I can’t have played it in fifteen or twenty years but I still think about Tokyo Megaplex every month or two. It tapped into something deep in my brain.
Thanks for the reminder of it. I won’t know what heaven looks like until I’m at its gates, but when I die, I hope I get to ascend via that weird updraft tube, and ride those red and white candy cane rails for eternity. If not, God might have missed a trick. What an incredible piece of work that game is.
I think one of the beautiful things about America is its formation as a country contains so many nations, and in a way America’s cultural history is inseparable from those nations it contains (newly formed) extensions of. America is a young country made from the offshoots of old countries. So in some ways, I think America’s Homer can simply be Homer, just as America’s Shakespeare can be - if an American chooses - Shakespeare. That’s a beautiful idea for a nation.
To think about the question less evasively, for me, it depends on which aspect of Homer is primary. I’m restricting myself here to widely read texts which I think have had vast impact on national character.
If it’s the idea of suffering, frontier, and the impossible but attempted journey, I find it hard to look beyond Melville. Not just for Moby Dick, but I would argue Bartleby, the Scrivener is an incredibly prophetic work about how regular people would come to experience the deep strangeness of modern American life.
If it’s a question of language - of how we write and speak to each other - so much of American language rests on Hemingway. This answer is unavoidable for me. In American communication, so much value is placed on clarity, and brevity, and the idea of “cool”.
If it’s Homer as an epic, sweet, sad, narratively-loose, arguably-multi-authored, and kind-of-orally-told document of American life, widely shared: Why not Homer Simpson? The Simpsons is art at its best, and has clear export and staying power for audiences.
I think the real answer is probably that it’s too early to say. Bob Dylan will probably be on this list in a century - but I think the ways his work is prophetic of future art and the broader future world have not quite formed into reality yet.
Scott Fitzgerald might be there if America’s 21st century experience is so defined by striving, (self) deception, and startlingly violent tragedy as its 20th century was.
Similarly, if America stays as strange and psychedelic and new as it has been in the 20th century, a case could be made for H P Lovecraft.
While I don’t personally enjoy Lovecraft’s prose, his ideas are increasingly prescient. In the 20th century, Americans walked on the moon. In the 21st, Americans have invented - in the consumer LLM - a thing for which the most tractable analogy is Cthulu, with its many-faces of a single entity, its cultists, and its nondeterministic and unknowable - or incomprehensible - non-human motivation or end-point, which can only really be described as a desire for its own momentum.
It's a product that's so deeply dependent on lifestyle, and on physical needs. My fiancee is technically partially sighted (nystagmus from albinism) and I was surprised to learn when we started dating that the RNIB has 3% of the UK living with some form of sight loss, and ~350,000 people as partially-sighted or blind. I didn't really understand the full and nuanced spectrum of what "sight loss" could mean - I guess I had thought of it in very cartoonish, uninterrogated "no glasses vs wears glasses vs sunglasses-dog-and-cane" terms.
The age demographics which read more, broadly speaking, have vastly higher incidences of some kind of sight loss. It's a sizeable portion of the market that goes, it seems, rather underserved in products akin to e-readers, like televisions.
(Televisions specifically being a product where - somewhat off-topic - there are an awful lot of features that would be trivial to implement which would drastically improve the product for the meaningful segments of the market who have some form of sight loss. These features unfortunately seem often to be lower down the roadmap than spyware-adjacent tracking or new kinds of motion smoothing for users to turn off.)
My own IRL library is scattered in storage crates across places I've lived in the past, due to space and travel limitations, so I strongly empathise with you on that.
The creepypasta operates as such an interesting cultural object: Whilst having an original - often anonymous - author, a creepypasta is often built upon quickly and communally, making it strongly akin to a folk tale that develops at a previously impossible speed.
To be successful, the original generally needs to contain a highly compelling image (linguistic or graphical): the "backrooms" in this instance, the long limbs of Slenderman. I would personally guess that the increased cultural interest in the Wendigo over the past few years is directly related to the "Anansi's Goatman Story" greentext's amazingly sensory description of the smell, which really stayed with me. [1]
These alleged actions from A24 are anathema both to the way these stories develop - communally - and also to the communal, somewhat anti-corporate, "independent" creative values in certain audience demographics that partially contributed to A24's rise in the first place.
I suspect it's just copyright strikes being applied thoughtlessly, in the usual way, without the legal team (or whatever is automating their actions) realising that this is an instance where an exception should be made.
To me personally, even if it's not maliciously intended, it's still gross, or perhaps even grosser.
Incidentally, I thought the film itself - while relatively unfaithful to the source material - was one of the better attempts I've seen at integrating the ugliness, slop, and grotesque aspects of online life into a contemporary film. Usually, the online world is translated much more clumsily.
This looks nice - I think ‘small enough to be pocketable’ is an important form factor. Reading preferences are a very personal thing, and for people like me - who feel they internalize more effectively when reading from a physical book better than when reading from a screen - this form factor makes owning one more worthwhile.
I have always wondered why the form factor that has been settled on for e-readers was ‘book-like’, other than the obvious screen-size advantages. I wonder if there is a better form-factor out there and if anyone has any ideas. Electronic devices don’t need to resemble the thing they are replacing, and it’s sometimes better that they don’t.
I know a lot of people lust after the clamshell e-reader imagined in the film It Follows. [1]
The haptics of a physical book are, I think, what helps me better remember them, because memory is intrinsically linked non-digital sensory information like touch, and smell.
I’ve personally tried a few prototypes to try to bring unique sensory inputs to individual books read digitally to help me remember each one better: converting an AliExpress Game Boy into one to enable more haptic and visual differentiation; generative ambient & binaural music that is automatically created based on the text shown.
Each did help, to an extent, in preventing the way everything I read on an e-reader slightly blurs into indistinct memories of ‘reading generally’, rather than ‘reading this specific book’.
Maybe the e-reader has to be as personally customizable as a cyberdeck for those of us who kind of need the other sensory inputs. So it’s good that the firmware on this one seems to be open-source, but I haven’t yet read through it to understand the extent of that.
twitter.com/gabriel666smith
henrygabrielsmith at gmail