7,479 karma · joined March 14, 2016
[1] https://www.bbc.co.uk/news/topics/cp7r8vglgj1t
Would be great to have a version where the game assets are designed/generated somehow and it's already to be tested, even if the experience is worse.
I think each part is generally treated like an individual large program - which makes sense because otherwise you have large parts of the model sitting there doing nothing. You want all of your silicon running hot, RAM sitting doing nothing for a period of time is a waste.
> If OpenAI had needed a single addressing plane to cover all their users, 128-bit might be required. Similar to how the internet hobbles along fine with 32-bit addressing by just adding a layer of indirection to the internet with NAT.
I think even then, 128 bit memory for OpenAI et al would be more of a headache than it is worth. Economically it's not worth building the 128 bit systems. The fact that servers and consumer CPUs have a lot of re-use within their designs reduces the costs for everybody. Sony for example stopped producing their 128 bit CPU [1], where it was mostly about pushing more data between CPUs and VPUs. Now we have specifications like PCIe where you can just adjust your pipe width to increase throughput.
> I guess the other example of 128-bit addressing is ZFS. What are the biggest ZFS file systems? And who has more data? S3? Once you get bigger than 64-bit you want to scale out horizontally anyway and not just put it all into one flat addressable plane.
Yeah exactly. The other thing to consider is that it doesn't matter how many bits wide you can go, the bottleneck is the smallest width. With ZFS, at some point you likely need to send it over a network, and I think fibre is up to 64 bits via parallel transceivers. If that points outwards, it'll probably be single bit by the time it gets to you.
> The two parts of the antenna are intentionally designed to create an omnidirectional radiation pattern without a pronounced directionality.
I worked on a project a while back to try to utilize this to detect the arrival direction of WiFi signals approximately based on received strength. Do this for enough base stations and you have a very crude positioning system.
> Note that "minor" implementation issues like die space, routing, and gate delays, especially of 128-bit adders & shifters are non-trivial, so people aren't going to rush out and build 128-bitters for fun, just as people matched timing dates of their 64-bitters to their expected markets.
I think we're stuck with 64 bit for quite a while. The circuit size jump from 64 bit to 128 bit is significant. There's no fixed scalar or apples to apples comparison (that I'm aware of). Just for adders though, 2 bits requires 2 full adders, 4 bits requires 4 adders, 8 requires 8, etc.
Another way to look at this is that 16 bit gives addressable memory up to 65k, and quite a few programs had to deal with paging in architectures like the 8086/8088. 32 bits gave us up to 4GB addressable memory, and not it's now not uncommon that a program such as a web browser exceeds this. 64 bits would give us up to 18 exabytes of addressable memory. I'm not aware of any common programs breaking into the terabyte category (even in most research), let alone petabyte and then exabyte.
Even iterating over that many numbers becomes a large computational task. Just a quick test program:
// gcc -O3 count.c -o count
#include <stdint.h>
int main(){
uint64_t z = 0;
for(uint64_t i = 0; i < UINT64_MAX; i++) z += i;
return (int)(z % 2);
}
> The end result exactly matched my expectations—the best criterion for success that I can think of. With an LLM providing the heavy code lifting and the CAD work, I think I spent more time waiting on supplies to arrive than on anything else—something attributable to my lack of planning and the ease of next-day delivery.
I think the author robbed themselves of a deeper understanding.
[1] https://www.npl.co.uk/products-services/time-frequency/msf-r...
The issue with the crypto argument is that it is purely political, it is not affecting the service at all. You may think that you agree because it happens to align with your political sentiment, but it's entirely possible they come for your projects next.
> Allowing these projects is not a neutral stance; the effects to the environment are severe. The "DarkFi" (seemingly some crypto designed for criminals, citing an FBI director) project for example uses a "PoW blockchain" (per their readme). PoW (Proof-Of-Work) projects waste an astonishing amount of computational energy, with Bitcoin alone using more energy than the entire countries of Argentina or the Netherlands. This energy is almost always from fossil fuels, which directly contributes to greenhouse gas emissions and undermining any global efforts to fight climate change.
> This is not a neutral space, Codeberg is political. If you disagree with someone, be constructive and don't call them merely political and think that's a reason for the discussion to stop.
Cryptocurrency is bad, but LLMs are fine? What next? Maybe you can't store your simulation source there because it is deemed to resource intensive for the expected results? When did source code itself become dangerous?
It looks like we need more alternatives...
My plan is to not use this firmware and use MicroPython instead (based on a high-level OS I already developed that is extremely low battery consumption). I carry around with me an A6 notebook with TODOs for each day - just the other day I temporarily lost it and realized that I actually need to have a backup for this stuff synced with a private VPS.
I would look at swapping out the HDD for something solid state - lighter, less power, higher performance for random R/W.
Then it's just a case of lightening the load of the CPU as much as possible, strip out everything that is not needed.
I've run modern Firefox on much lighter devices. One of my netbooks is the same spec as this, and I do browsing + coding on it easily.
This is a solved problem though. Wireless earbuds can do it. We're probably just talking about a TVS diode.
> For a niche, low volume product with an unknown market demand I think making the simplest possible version of the product is a good idea to start, but at $99 it’s getting into the range where buyers don’t want to think of it as a disposable item.
> The bigger problem is that the 2 year battery life depends on the device being used for only short notes like “Add milk to the grocery list”. The people who expect to use this for taking notes or thinking out loud could exhaust the battery in a couple months.
$75 for a use once device that could last as little as a few months under normal usage, or $100 for something that could operate for 5+ years. Knowing whether there is a market is always difficult, but if you do crack a market you typically only get one chance to get people onboard.
> Roughly 12 to 15 hours of recording. On average, I use it 10-20 times per day to record 3-6 second thoughts. That's up to 2 years of usage.
2 years is optimistic I think. 3-6 second thoughts is not much, most ideas I'm having past TODOs are >15 seconds. Plus this would definitely depend also on how regularly you sync to the app.
Bare in mind, these devices have not yet been used in the wild on people outside of the product creator. Maybe one of the use cases is to note down monologues, or to easily record conversations with clients for later review.
1 month battery life is also highly usable, but there at least should be a way to charge it.
> Adding the port, the USB-C PD circuitry may take up too much room.
You don't need PD circuitry for this. 5V 500mA over USB2 is more than enough. But the actual port and battery charging circuit is a bit ridiculous at this size.
I'm not sure precisely what the EU rules are, but the requirement should really depend on the dimensions/power of the device.
> Why can't it be recharged?
> We considered this but decided not to for several reasons:
> You'd probably lose the charger before the battery runs out!
> Adding charge circuitry and including a charger would make the product larger and more expensive.
> You send it back to us to recycle.
I don't think this is true. The charging circuit could have been in the charger itself. To provide access to the battery, one of the terminals could be behind a transistor enabled by the micro. The charger could then send a signal to the ring to unlock the battery terminal. Then all you needed to do was expose two/three pads externally.
If you're looking for a good USB3 hub, look for one with a short thick USB cable, metal chassis. If it has HDMI it's a good since because you're unlikely to pump that via USB2.
WEIGHTS = {
'w_xg': 0.09,
'w_goals': -0.07,
'w_star': 0.018,
'w_value': 0.18,
'w_rank': 0.4,
'w_def': -0.12,
'xga_share': 0.85,
'w_gk': 0.0042
}
$ python3 worldcup_model.py --sims 100000
2026 FIFA World Cup -- championship probabilities (100,000 simulations, from Round of 32)
1. England 11.7% *
2. France 10.0% *
3. Spain 9.3% *
4. Argentina 8.4%
5. Germany 8.1%
Terrence's issue here is that he is over complicating things:
- Toothbrush - Just take a non-battery toothbrush, it's perfectly fine and will not break down or need charging.
- Tracker - I'm doubtful you will get it back in most countries. Always keep the essentials such as phone, wallet and passport on your physical person where possible.
- Bug Bite Zapper - Do as the locals do, they too don't want to be swamped with insects.
The point at which the metrics become meaningless is when others become aware of them, and begin to optimise for them. Lines per code is is not a bad insight for development activity, only when the developers are not aware of the metric. Price per 1M tokens became meaningless when LLM providers started to optimise for it. It seems to be that Sonnet 5 is optimised to score well on AA intelligence whilst seemingly having a lower price per 1M tokens.
I think generally we are in an AI bubble, and it will at some point pop. The numbers simply don't make sense. I would gamble heavily on local cost per task to survive the LLM winter. Given that hardware is pretty much a fixed overhead, you probably want to optimise for task per kW - that's where I'm betting.
I thought they came in different variants - my meat-based LLM must have hallucinated it.
> It's also the one everyone can buy online for $19, not Pi Zero 2 W (always out of stock)
I get the impression that they don't actively manufacture Pi Zero W anymore and it's all old stock. Rapid for example say they have an availability of 761 (at time of writing) [1], but are those all within Rapid's control, or what is reported at Raspberry Pi's remaining stock? If you sell 1k printer units, can you put a Pi Zero W in all of them?
Compute modules are probably a bit more reliable and there is some capability with other modules, specifically the DDR2 form factor ones [2] (which is my favorite form factor). For example, ClockWork sold their own compatible compute modules [3].
[1] https://www.rapidonline.com/raspberry-pi-sc0020-zero-w-singl...
[2] https://www.rapidonline.com/raspberry-pi-compute-module-cm3-...
A few other comments have concentrated on the printer head and ink cartridge, but the roll vs sheets interests me. Manipulating sheets of paper is actually quite a difficult problem to solve - and there is only a demonstration of paper placement. Loading page after page is actually really not easy at all. There is no example of it printing.
I see they got a nomination for a design award [1], which in my personal experience has been a negative signal for successful projects. On the same page they mention that they also mention that they don't even know how much this will cost yet:
> We know this is your most frequently asked question! Final pricing depends on a moving puzzle of production volumes, BoM costs, industrialization expenses, regulatory certifications, and final engineering developments.
On the crowd funding page it mentions for parts [2]:
> Main board: Raspberry Pi Zero W
> Cartridge board: STM32 MCU
Not a specific MCU variant, and the specifics of the Pi variant (i.e. RAM) are not mentioned. I would expect these parts would be locked in by now. I have a feeling that this is not quite as ready as made out - I think there is still R&D going on.
[1] https://www.crowdsupply.com/open-tools/open-printer/updates/...
> Too many people seem to believe that the AI bubble popping means something like "the total volume of the activity will precipitously drop and whole centers will be abandoned forever".
I think that many of them will fail. We have several data centers being built nearby to me, where the UK has some of the highest energy prices. I don't know how AI companies can make it make economic sense, let alone any normally operating company (like a Cloud service provider).
> I don't believe that. I was an adult when the dot-com bubble popped. What happened then was that a lot of unsound businesses went under, and the healthier players (like Amazon) expanded into the resulting void. Ultimately, the relentless march of digital technology didn't even slow down back then, although the inflow of risk capital definitely did. But the existing resources were mostly taken over by someone else.
I think the difference between this and the dot-com crash is enormous. Every part of our economy is now entangled in this. A significant part of the US's GDP is based on the idea that Nvidia can keep expanding their market cap past several Western nations. Your pension likely has large buy-ins on the AI hype. Everybody at every level is very exposed to this.
> If/when the current bubble bursts, it will be similar. All that infrastructure will be resold and re-rented to healthier players. By now, machine learning has reached a level of usability/maturity which is genuinely useful and the economy at large won't abandon it any more than it abandoned e-mail or WWW back then.
I have my doubts that anybody will be liquid enough to buy what remains at reasonable prices. And the impact of that is obvious, why buy RAM from Micron when LLM Corp are selling it for 50% of retail value?
> When children are very young, parents can set strict boundaries. But as kids move into their teens, parents also have an obligation to loosen them. Teenagers need spaces where they can act independently, make decisions, and talk to others — where they can learn to navigate the world without a parent looking over their shoulder.
You are still responsible. If you allow your child more freedoms, you are still responsible if something bad happens. Your choice to loosen your parenting shouldn't become everybody else's problem. Some parents for example allow children to try a small quantity of alcohol - if your child is suddenly found drunk, they shouldn't ban alcohol for sale everywhere.
I would go the other way, if it's found as a parent that your child is drinking, smoking, etc, then this carries punishments for the parents. I think it should also be punished if they are found accessing online services targeted at adults.
> The digital version is the same idea, with cryptography. An authorized issuer verifies your age once and issues a signed credential:
Sounds great, but there is a clear agenda for digital IDs that 'they' are trying to shoehorn in with this "protect the kids" thing. They tried rolling it out digital IDs in the UK in 2006 [1].
As I said the other day, what really makes me suspicious is that most Western countries suddenly have the same idea at the same time. This isn't just some random politician wanting to protect children, this is an international concerted joint effort to roll out a form of mass censorship.
> So instead of fighting a system that is built to preserve privacy, we should be fighting to put the right checks in place — the ones that guarantee the implementation actually honors it.
The objective is the system itself. Once there is a control in place to stop you accessing these services/systems, changing the exact unlocking procedure is trivial and can be done gradually. We'll all just be frogs in slowly boiling pots.
1. Some tools take time to get setup in the way that they were originally setup. I mentally want everything to remain where I left it. I'm usually working on 5+ projects at a time, it's already enough mental tax to have to context switch, let alone to have to go find all the resources I need to be productive again.
2. I'm forced to use Ubuntu, and it is constantly updating stuff and making it crash. Every time I reboot it's a gamble as to whether something I previously used works again properly. Maybe 1 in 100 boots, the entire system doesn't come back properly.
This is some next level gaslighting. AI companies significantly increased the demand and have deeper pockets than most consumers. He who bids most gets the product.
> Mehrotra also said the memory shortage could last into 2027 and beyond because fabrication plants take years to build, and the next generation of memory is even more complex to manufacture. Micron is now investing up to $200 billion in manufacturing and R&D, including new memory fabs in Boise, Idaho, and Syracuse, New York.
When the AI bubble pops, and supply significantly outpaces demand, we'll then see them close tonnes of fabs.