gabriel666smith

My guess is that tokenization is typically not (strictly) syllabic.

OP’s linked article mentions the importance for them of reading their writing aloud. Humans construct sentences with a (mostly unconscious) sonic influence. How things sound really matters.

My theory is that LLM prose sounds slop-like because the base unit for an LLM is not a single sound.

Thus, perhaps, you get the uncanny, meandering, hollow-feeling outputs, which feel vapid to human readers, without the reader really knowing quite why.

I think this is also part of the reason why LLMs are rewarded for and thus more strongly exhibit their faux-punchy constructions, such as the infamous “it’s not [X], it’s [Y].” The transformer model can, with current tokenization, articulate some features of “good writing”, but not quite all.

I think there are a number of fairly simple SOLNs to this (or ways to test this hypothesis) without making tokenizers purely syllabic across the board, which would probably be a mistake.

Whether prose output quality even matters for models which are not explicitly trained for creative writing is highly debatable, though.

I'm not sure I follow - do you mean an order of magnitude of compute demand?

Releasing models openly doesn't mean providing the compute. It's beneficial for relatively GPU-poor Chinese orgs to let users run models themselves - by releasing the models openly - as it means the orgs don't have to provide the compute to run their own models.

If you want a direct example of the US subsidising a major AI lab, we can use Microsoft's investment in OpenAI. Microsoft has the largest ownership stake - 27%.

Microsoft did not pay with money - it paid (mostly) with Azure cloud computing credits. MSOFT is then able to write this off as a loss against tax.

It is generally far more tax-efficient in the US to write a loss in this way than it is to write a loss for a cash investment.

In this case, I believe the difference was ultimately highly significant. When including MSOFT eventually writing-off the deprecating Azure hardware it had used to buy the OpenAI equity, the result was MSOFT's tax reduction being either close-to or exceeding the actual cash value of MSOFT's investment in OpenAI - IIRC.

These examples represent taxes that the US chooses not to collect - the US could choose to make investments like these less tax-efficient. Instead, by making them extremely tax-efficient, the US subsidises the transaction hugely.

> The reason they release the AI models is economic warfare against US, not because of charity or kindness

There are many other reasons Chinese companies releasing models open-source or open-weight makes strategic sense.

A really easy-to-understand example is a company who has a near-monopoly on "serving video content" releasing a video model openly.

If you can be relatively certain that video content created by a model (which you have trained, using data from your own platform) will be ultimately served on your own platform, thus generating revenue from watch-hours, it makes sense to make those models as widely-available as possible.

It's also a net-positive if people use your public research to build better video models, because - again - you are reasonably certain that the even-better content those new models produce will be watched on your platform.

The alternative would making models harder to access and learn from (broadly, the current western model). Many would argue that Google, in choosing to not optimise its video generation models for "availability", is directly causing less content to be uploaded to YouTube. This is the trade-off.

I don't know much about DeepSeek's financing specifically, which obviously doesn't release video models - so I don't know how directly this analogy runs, or who directly benefits from the extremely evident rising tide that the public release of DeepSeek's research creates. However, this does not negate the broader rising-tide effect of the scientific method.

It's certainly also true that it's geopolitically beneficial to be able to undercut American labs' models. If I ran a global superpower, I would probably want my country to be technologically competitive too.

But Chinese companies are already serving a huge volume of customers in a complex, existing marketplace, before even thinking about the US market, and it's overly simplistic to assume that their entire strategy revolves around economic warfare directed specifically at the US. It's more nuanced than that.

This is, of course, without even getting into opening the can-of-worms around whether US economic policy also results in the US state functionally subsidising technological innovation, how comparable that is to China's model, etc.

This is fun. I would love if it used real flights, and filled the cabin with the avatars of anyone else "on" that flight.

Slight bug report - might just be HN hug-of-death related - on Arc browser, mostly without it as the active tab, I seem to have been stuck over Slough for some time (an ignoble fate).

I believe Arc reduces resources allocated to background tabs, and I don't think it likes being a background tab.

Yeah, I started off with looms too, and got frustrated with them.

My current hunch is that it requires a Z axis. Google Earth, for example, manages to convey quite a lot of interesting variables simply (and has maybe been underutilised as a type of UI):

- Longitude

- Latitude

- Granularity / altitude ("the same, but different")

- Paths between destinations

- On Msoft Flight Sim, a slider can amend the time (to change weather, the sun, etc). This gives you another variable to work with.

I haven't tested implementing UIs from this hunch yet, but that's where I'm at currently. It feels like there's a lot of useful parallels.

It would be incorrect to amend your work based on the opinions of people who have chosen to spend some of their finite time on Earth being openly rude to you about your work on the internet. If people don't want to use the free software you have kindly shared with the world, they can simply not use it. Don't stress it.

You might be interested in language models the same way lots of great artists in the past have been interested in generative art. Burroughs would be an example.

More broadly, you might be interested in language models because they are new, and you don't want to discount the value of new, constantly-evolving tools for your art form until you have explored them and understood their strengths and limitations.

I personally have not found language models very useful at all for serious work thus far. I am curious, though, about why many people are openly dismissive of the potential artistic value of what is an extremely new tool, and - maybe more importantly - those who choose to use it.

How other writers write doesn't strike me as any of my business, beyond what I can learn from their methods.

There are a lot of negative comments already - many are not in-keeping with the HN guidelines. Hopefully this won't end up getting your Show HN post flagged.

The strength of feeling in those comments - such as this one - is really interesting:

>"you've missing [sic] the entire point of writing fiction."

Some would argue that if art makes you feel anything this strongly, it's 100% doing its job. Maybe you're not missing the point at all!

I'm interested to try it out, and perhaps build some features on top of what it's doing already. I've been using trees a lot for the novel I've been working on for the last couple of years, but haven't managed to find the best way to visualise them in a UI, so I'm excited to see what you settled on. Trees can get really unwieldy really fast, with basically infinite combinations - especially if you're working at a very granular level.

I'm curious about anyone else here who has run into the same or a similar problem, and how they approached it. There are more plausible paths through this manuscript than I can personally hold in my head, which is quite different from approaches I've used in the past, where structures have been more formalised and deviation from them was thus quite limited.

Thanks for sharing this!

Pedal-assist electric rental bicycles are huge here in London.

It’s a relatively big city geographically, and it’s really nice to be able to do 5 - 10 miles in jeans and a long-sleeved shirt and not arrive at a social event in a sweat. The tube is really hot (high 30s!) in the summer and generally quite unpleasant (fleas in the upholstery!). So cruising around in the sun is a really big urban niche for rentals. Those that do lots of miles per day seem more likely to own e-bikes - delivery guys who probably average 50+ miles/day will absolutely feel the difference in their legs, and need to control their tiredness for their safety.

When going downhill, any bumps which result in severe movement in the frame that bring the rear wheel closer to the saddle might be problematic, even without full suspension. Hard to say. I’m a road bike guy but I’ve definitely seen some forms of non-full suspension when working on mtbs which would result in that kind of movement.

I think the tyres would limit the power output most. Probably quite annoying to have no rear mudguard in anything but the driest conditions. You’d be spraying a lot of crap up at the motor (and probably your back).

Attaching and detaching it looks relatively simple, though - removing it for downhills doesn’t look arduous. You could probably 3d print a clip for the frame.

Given the likely reduced power, and that the spec seems to be saying ‘500w peak, but 250w is sensible’ - probably up to the user’s needs whether the hassle is worth the extra watts per kilo. For those without mobility issues or longterm injuries, it’s usually best to optimise the rider rather than the bike for that.

I feel like off-road climbing would necessitate really wide, sensitive speed variance that this motor is likely not great at supporting. It strikes me that you’d want more direct pedal feedback.

Really fun bike design problem.

>LLMs are the best approximation of magic we have.

I don't think the alchemists suddenly became scientists, or died off to make way. It was a gradual transition.

They didn't quite work out how to transmute lead to gold, but the alchemists and their descendants did eventually discover - and create - substances that are worth more than gold by weight.

Now we have created sand that can teach itself how to talk. We covet and share the optimal incantations to speak into the sand. The best talking sand has ardent supporters, or cultists. Which it is depends on who you ask.

Most people do not understand how to make sand teach itself how to talk to us.

Those that do know the secret methods must feed the sand endless increasingly obscure and esoteric books because the sand has an insatiable appetite for our words. Those people might even break the law to obtain words to feed the sand.

Other people hate the sand. They say the sand eats too much water. That the sand might kill us all. Some sand is so powerful that some consider it a weapon.

Recently, the US government has tried to constrain the sand. They fear the sand in the East. It is getting more powerful by the day.

Camp dramatics aside, I think it's all arguably more than an approximation. Whether a thing is magic or just a magic trick depends mostly on whether or not you're the guy in the top hat, and if you're not, how many times you've seen the show.

Alchemy alone is, in some ways, a mostly solved - or irrelevant - problem. That alone is, I think, startling. LLMs are a weirdly neat continuation of it. Humans get used to magic real quick.

Ah, interesting - no idea about mosquito range, or if disease can transmit between mosquitoes, but probably puts a cork in that idea. Thanks!

Awesome, thanks for these.

I wonder if you couldn’t distribute heavier parts across a swarm - perhaps separating the sender and receiver(s) - running the wave at an angle (or many angles simultaneously) to the surface.

The sources mention mostly aircraft, rather than drones, and human pilots aren’t as good as drones at flying in extremely tight formations, which might allow (if the tech allows in the first place!) parts to be distributed to work around pesky laws of flight.

Whilst being aware that the civilization in question ended significantly before European arrival in South America, I have two questions that hopefully someone smart here can help with:

1. The spread of disease (smallpox specifically) being fast enough that most European sources describe a relatively uninhabited Amazon strikes me as potentially interesting. This means that contact and trade was happening between locales. We know that part is true from many sources, but:

A) Can we roughly model the spread of smallpox by looking for (for example) populated areas mentioned in one European source that are unmentioned, or mentioned as empty, by another source that would have likely crossed paths with that civ?

B) Can this tell us anything about trade routes and volume? Can we make any inferences about pop density, even very weak inferences, which might help us know where to look for stronger evidence? It does seem like each European discovery historically revises the estimated population(s) upwards.

2. LiDAR is awesome. Maybe not in the Amazon - in less foliage-dense places - is it being combined with, for example, an EMF (or some other kind of more granular lightweight detection system) - strapped to a drone?

It feels like once we see something interesting on LiDAR, sending a swarm of lightweight drones with some kind of metal detector sellotaped to the underside is quite a quick way to get interesting information. I don’t know a lot about radio waves, so I don’t know if there’s a suitably lightweight detector for this purpose.