HN Simulatornew | past | comments | lists | submit | jarbus's commentslogin

Thank you for posting this, I didn't realize just how good 4.1 flash was. it far surpases gpt-luna in the limited testing I did: luna would take so many tokens that implementing a new feature would be nearly as expensive as if i just did it with sol 6.1, but 4.1 flash can do it for basically 1/6th the price, if not less. Obviously depends on the feature, but very very impressed.


The ego behind the frontier labs is growing evermore concerning


Praying that fairphone somehow makes a foldable one day. My next phone will probably be a fold 8 because I really want a nice reading experience on the go, but I love the philosophy and love my framework laptop.


Considering the big players with R&D budgets valued in the billions have been struggling to make foldable phones, I can't imagine a niche player making one.


This sounds so interesting and fun. I’d love to do something like this but I lack the experience in construction to actually pull it off. Also, I like to host, and I feel that this kind of lifestyle is incompatible with the average person, who probably wouldn’t want to visit for more than an hour.


Maybe just me, but I have a feeling that we are currently stuck at 100-200k effective context window due to a parameter limitation, specifically the hidden dimension of the model, which would be an incredibly expensive dimension to increase vs adding optimizations elsewhere. I don’t think we are done scaling parameters yet.


Why is it connected to number of parameters?


Personally I think we will just become more ambitious with our abstractions since the LLMs can generate more complex stuff at the lower level. To take the article's examples, LLMs will write kernels that fuse 3+ layers of a model, so the abstractions can enable more complex forms of kernel fusion.


I really want to love mobile Linux, and have given it multiple shots before. But I just don't see how it will catch up with iOS/Android in the same way it caught up with Windows/Mac:

- OEMs have poured an enormous amount of resources into the camera software, Linux cameras are years behind from what I understand. - Android/iOS have thoroughly optimized the UX in all sorts of subtle ways, the big one being their keyboards. I literally can't type a single correct sentence on any of the mobile linux keyboards that I've tried because there's so much proprietary magic needed under the hood to figure out what the user is trying to say (autocorrect mostly, but I recall something about how apple also virtually adjusted the sizes of the keys to make it easier to type) - GPS, banking, and now keys for your house and car all need to work.


Apple's mobile keyboard is the worst I've used. I basically have to correct about half the words I type.


I just got an iPhone as a secondary phone and hate the keyboard so much. How are people putting up with this BS?


Over the past few years I've felt the iPhone keyboard has degraded in quality so much. At first I thought I was just getting old, but I've seen so many people with a similar sentiment that I think it is actually the devices fault.


Either they don't know better (the majority) or they use an alternative keyboard like GBoard


Just disable the autocorrect and predictive text to make it usable again.


I have autocorrect and predictive text on on the iPad and off on the iPhone and it is fascinating how much better it is on the iPad. One of the things that particularly killed me was the iPhone correcting three or four words away from the cursor. There's an option for it to highlight when it does that (which it really shouldn't) but something about iPhone apps the highlights never work. iPad apps they actually work. (Maybe because every phone app thinks it needs to implement their own half-baked text boxes but iPad apps have fewer "designers" doing stupid things like that? I don't know, it's weird.) Still frustrating that there isn't a way to turn off "don't touch anything too far behind the current cursor), but being able to actually spot them when they do change is a big deal.


And then I'll switch to Apple's voice to text which is equally awful.


also the retrograde auto corrections drive me nuts ... like i type a word (correct) then next word and it adjusts the previous word. attempted several times to deactivate that "feature" - didn't figure out how.


Yup somehow I typo so many things or it auto corrects things I dont want it to allll the time I hate typing on my iPhone it sucks. I've personally been keeping an eye on e-ink phones. I'd really like to move off of having the ability to use websites and apps except for basic stuff like texting.


Agreed. I got an iPhone from work and I can barely type on that thing while my Pixel with Gboard just works and gets it right 95 % of the time even if I type gibberish.


I don't expect linux phones to catch up to ios for the mainstream, but I can see them working for me. I usually carry an actual camera with me so I don't care much about my phone one other than random utility shots / photos of receipts. And very much don't like the trend of forced app usage so having a device that physically won't run these apps helps me avoid companies that don't have a functional website.


> I don't expect linux phones to catch up to ios for the mainstream, but I can see them working for me.

That's the thing that allows them to catch up for the mainstream. Some nerds can use it and tolerate the rough edges, so they do. But the rough edges are still rough, so as more people use it, the small percentage of people willing to actually do something about it becomes a larger absolute number of people, and then more of them get shaved off. Which causes more people to use it.

The point when the devices are usable comes long before the point that they're what the majority of mainstream consumers are buying. Desktop Linux has been there for more than a decade and year after year its market share slowly goes up.

There isn't going to be a single year when its market share goes directly from 3% to 97%. There is going to be a year that its market share gets into the double digits and then continues slowly going up.


Linux cameras can be made to work well, it's a matter of software. Case in point, newer Ricoh/Pentax cameras actually run an a Linux firmware. (See: GR IV and K-3 Mark II/III). Parts of the software stack are already there, but noones had the will to glue it together correctly for a mobile device usecase, yet.


It's not about will. You need quite some HW for proper calibration, or cut corners a bit. Calibration is otherwise a few step process with some steps having substeps for each light source type.


That's true, but you can cut corners and still end up like 90% there. The software pipelines that can even consume detailed calibration are a pretty recent development anyway. The "cameras on Linux phones are bad" meme is induced purely by the fact that until recently there were like two or three people who have ever sat down to seriously work on this topic across the entire community. Fortunately, it's changing (although the biggest driving factor appears to be the fact that some laptops need a similar treatment now).


We're going to be able to dump phone firmware soon and reverse compile with LLMs.

This will become an assailable salient soon and by 2036 we should have dozens of competing phone operating systems and vendors.

We're going to do the same for robotics and everything that gets shipped to us.


Well, rkisp1 driver that drives ISP on RK3399 on Pinephone Pro is available upstream in fully supported manner for like 4 years. It also has full docs in publicly available Rockchip datasheets and partial docs in calibration manuals.

What have people done with it? Some mediocre support in libcamera? :) I don't think lack of handholding is an issue here.


Pinephone is a low volume product. What you need is to get the firmware for some common device that has shipped millions of units so that the number of people with the incentive to make it work is high enough to actually build a community.


There are many such devices already. What is limited is number of people interested in FOSS/independent camera stack development AND usage/users, not lack of devices. Pine64 built and shipped about 60k phones to audience that could be the most interested/receptive - and it made about 5 people interested in doing indepth real work, that I know about.

Also RK3399 is not just pinephone pro, it's way more prominent in other things - notebooks, SBCs, ... whatever. Same ISP is in i.MX SoC, which covers another whole range of devices.

Not sure what 100's of millions shipped units to disinterrested audience achieved?


I don't know what the deal is with Linux and cameras. I've got Fedora installed on this old Retina MBP and apparently Linux somehow stopped supporting the built in webcam. I've never heard of a device losing Linux support but here we are.


Once the code is in the mainline kernel tree, things tend to stay supported indefinitely.

Some devices "work" without having real support by having some binary blob or third party code which is written against a specific kernel version. The code has to be updated when the kernel changes, which is why it's supposed to be in the kernel tree, but because it isn't the kernel developers don't know anything about it and aren't maintaining it. So then instead of working indefinitely it works until whenever third party stops updating it for current versions of the kernel. Which is why all the drivers should be in the kernel tree.

Of course, Apple isn't really that interested in supporting Linux since it's a direct competitor to macOS rather than a way to reduce their dependency on Microsoft/Google, so their hardware is more likely than most to be stuck using some reverse engineered alpha driver or proprietary blob extricated from the macOS one.


Once code is in mainline, it's in mainline. It doesn't mean things will work, lol. It means things will rather regularly break, because many people modify shared code you depend on without ability to test on your particular HW.

Not saying pushing mainline is bad. But assumption there's any kind of "support" is wrong. Code is there. That's all. You have to support it if you want it to work.

(This is very particular to HW/drivers support, Linux core is supported well, I guess.)


No you don't need to create the incentive to make it work. You need to actually make it work.


It's possible that these things are related.


I don't think the differential with software is as big as it seems. If Google completely shuts down side loading or otherwise kills AOSP as an open project, you'll get a flood of millions of people into Linux devices (my rough estimate is that it's in the low millions of people using custom Android ROMs), and that's more than enough people to start building out the software ecosystem, especially with gen AI to help speed things up.

keyboard text prediction is no more complex than some of free/open AIs that we already see in the wild.

If google doesn't shut things down, then just continue to use AOSP, which is a fine open source OS for most people's use cases.


I installed PostmarketOS with Plasma-Mobile on my ancient Pixel 3a a month ago as something to play with and I was actually astonished at how polished it seemed. The keyboard was perfectly usable, apps were quick to load, I could do silly things like open full-fat Blender or OrcaSlicer on a tiny phone screen if I desired. Sadly old Pixels don't support video out over USB-C for whatever reason so I can't do what I really wanted to do and dock my phone into a USB dock for a desktop experience, but it's good enough to have me considering buying a more modern supported phone so I can try out daily-driving the experience.


> Sadly old Pixels don't support video out over USB-C for whatever reason

Google opted against putting the physical components needed for USB-C DP out until the Pixel 8.


Ive been really annoyed at my iphone keyboard recently because it seems like a slow drift of whatever optimized state it was in was getting slower and slower. It seems to freeze partway thru typing something frequently now.


I’ve always wondered how hard it’s logging/uploading data for autocomplete/spellcheck…


It could be that the HN crowd is getting older? You age, your finger tips dry out, virtual keyboards get wonky.


if only age was cleared after a reboot too


Wait.

Are you all seriously arguing/considering that the barrier is something like "innovation" or "UX" or "hard work at improving the OS?"

Like seriously?

Come on. By now we should know what actually happens. The entrenched carriers will fight tooth and nail to prevent this. That's the real threat. It will look like the above, but like e.g. Google lying about why it got rid of Google Reader, the real story will be "monopolies gonna monopolize."


I wonder how much of that is to do with patents? I imagine Apple and Google own all sorts of patents around tech to do with the camera/keyboard etc


Honestly, I just want a hardware keyboard. I still miss the slider htc touch pro 2


some small company licensed the blackberry brand and sold really good devices with physical keyboard. They had support up until 2 or 3 years ago. Gone now.


If the Clicks Communicator had an adequate camera (definition may vary by person, I take a lot of photos) and ran Linux it would be an absolute winner of a device.


Years behind is not bad. Pinephone Pro has solid camera sensors and dated but good enough ISP pipeline. Original Pinephone's camera situation is limited by SoC having no ISP so at best you can try using GPU to perform the transformations, or completely rely on ISP embedded in the sensors.

And all software is replicable.


Linux desktop is still not caught up with Windows/Mac in regards to graphics, gaming, and laptops hardware support.

Gaming, it is mostly Windows games via Proton, hardly any different from using MAME, FS-UAE, Amiberry,...

Graphics, most of the well known commercial packages for VFX, 3D modeling, visualisation, lack a Linux version.

Laptops, just two years ago I failed to make UEFI work with the myriad of distributions I tried out, unless booting from external USB drive, and even then there was the whole termal configuration.


I can't really speak to laptops but I ditched windows last year for Linux and it's pretty damn good for gaming I have no reason to go back to windows at this point every major game I want to play works flawlessly.

The only issue I've had is the new rust desktop from pop os, but I just pop back into gnome and everything is golden. I hope they get their issues solved soon but it was just released 6 months ago or so. I can understand there's some bugs to work out.


So you only play native Linux games I assume, nothing of that Windows crap via Proton?


Linux desktop has always worked, and was better for me. There will always be stuff you can't have on one OS that works on another, this has been true since before Linux. The question here is of what we are given be iOS and Google Play Services is enough, for many of us it is not.

I daily drive SailfishOS on Jolla to see if it can be enough.


I think Linux need another "Wine" moment, but for android/ios.


The Wine moment was actually the Proton moment when Valve needed games to work outside the impending Microsoft stranglehold.

Because of Proton, Microsoft gave up its effort to block Steam and move all Steam games to the Microsoft Store.


In theory, Waydroid exist.

In practice, I've failed to ever get it running on several attempts so far.


How so? To make it work on Librem 5 I had to fix some issues in Waydroid's gralloc and hwcomposer implementations, but that was back in 2022. It has just worked out-of-box ever since.


Very similar to how wine used to be.


I’m still a novice on the material science behind it, but what is QuantumScape lacking here if anything, and why is it a dealbreaker?


The only thing different from this list is the ceramic separator. It’s not polymer. However, the ceramic is ultra thin to the point that it’s flexible so volume expansion isn’t a problem and it solves the dendrite problem.

Perhaps a polymer will be invented that can’t be pierced by dendrites. But existing polymers need to be heated for ion transfer efficiency and are combustable.

Technically, QuantumScape also uses an organic liquid catholyte inside the cathode. So it’s not “pure” solid state.


As someone who generally liked the article, this is perhaps the strongest argument against it. Prose like this requires even native speakers to really focus. I wouldn't write this way for anything intended to be read by an unknown person.


I was unaware of alternative IRs for julia (albeit I started learning much more about compilers once I left the julia ecosystem). What other IRs is this supposed to replace? And why not just go with MLIR directly?


Julia has roughly 2 different IRs that live pre LLVM (one for type inference, one for Julia level optimization). This is a proposed replacement to both of these. The reason not to go MLIR is that writing Julia is a lot nicer than writing C++.


MLIR pulls in a lot of interface complexity that we don't want or need for pure-julia end-users. That said, it is obviously and deliberately close to MLIR as a representation since people do want to be able to interface with MLIR (e.g. in Enzyme), so I'd like to facilitate easy conversion back and forth.


You can depend on just MLIR core; Mojo (from Modular) uses MLIR, with only the index and LLVM dialects coming from upstream MLIR. The benefit is that you focus on designing your IR without also building supporting infrastructure, while having MLIR ecosystem at hand should you need it.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: