gemini 2.0 flash is and was a godsend for many small tasks and ocr.
There needs to be a greater distinction between models used for human chat, programming agents, and software-integration - where at least we benefitted from gemini flash models.
Seems like their whole business model was based on the fact that tailwind was difficult to use, and now with llm we have a simple way to use it in a good-enough way.
They, and other companies, should rather depend on corporate users. Don't let multi-billion revenue companies use your tech for free.
Seems like many companies leaned it a bit late, we always have the same news every fewe years (docker, mongodb, terraform, elastic).
> Seems like their whole business model was based on the fact that tailwind was difficult to use
Uhhh no... People already struggle with CSS. No one would use Tailwind if it made it even more difficult. I've used and loved Tailwind for 5 years and some without ever having any components written for me. At worst it's as difficult as CSS (centering a div is not any easier, you just write it in a different place), and in some areas like responsiveness (media queries like screen size breakpoints) the syntax is way easier to read and write.
The problem their business model was solving is first that good design is hard, and second that even if you can design something that looks good, you might not be good at implementing it in CSS. They did those things for you, and you can copy-paste it straight into your app with a single block of code thanks to Tailwind.
You're right that LLMs essentially solved this same issue in a more flexible way that most people would prefer, and it's just one feature of many.
Nah. Plenty people struggles with the use of tailwind or at least were interested in shortcuts. Thats the whole what tailwind plus offers. In some ways tailwind is like matplotlib/pandas/numpy. Increadibly powerfull but some methods/classes are difficult to remember to you keep googleing the same things.
Doesn't matter anyways wether their customers are people who search for shortcuts or people who search for "the best designs".
Their problem was and is that tailwind is used by many of the most profitable companies in the world for free.
Thats so unbelievable stupid. You have corporations paying millions for MS 365 subscriptions, confluence, and other software and basically nothing for a totally optional ui library. If the use of tailwind saves 10 engineering hours per month then it's worth it to pay a few hundred $ for a licence.
Given that their team isn't big they don't even need that many customers. Add a bit consulting for a decent hourly rate and they should be golden.
The more I think about it the more I blame the CEO for poor decisions.
What actually makes it an AI platform? Some tight integration of an intel ARC GPU, similar to the Apple M series processors?
They claim 2-5x performance for soem AI workloads. But aren't they still limited by memory? The same limitation as always in consumer hardware?
I don't think it matters much if you're limited by a nvidia gpu with ~max 16gb or some new intel processor with similar memory.
Nice to have more options though. Kinda wish the intel arc gpu would be developed into an alternative for self hosted LLMs. 70b models can be quite good but still difficult / slow to use self-hosted.
These processors have NPU (Neural Processing Unit) which is supposed to accelerate some small local neural networks. Nvidia RTX GPUs have much more powerful NPUs, so it's more about laptops without discrete GPU.
And as far as I can see, it's a total waste of silicon. Anything running in it will anyway be so underpowered that it doesn't matter. It'd be better to dedicate the transistors to the GPU.
The latest Ryzen mobile CPU line didn't improve performance compared to its predecessor (the integrated GPU is actually worse), and I think the NPU is to blame.
If you ask NVIDIA, inference should always run on the GPU. If you ask anybody else designing chips for consumer devices, they say there's a benefit to having a low-power NPU that's separate from the GPU.
Okay, yeah, and those manufacturers’ opinions are both obvious reflections of market position independent of the merits, what do people who actually run inference say?
(Also, the NPUs usually aren't any more separate from the GPU than tensor cores are separate from an Nvidia GPU, they are integrated with the CPU and iGPU.)
If you're running an LLM there's a benefit in shifting prompt pre-processing to the NPU. More generally, anything that's memory-throughput limited should stay on the GPU, while the NPU can aid compute-limited tasks to at least some extent.
The general problem with NPUs for memory-limited tasks is either that the throughput available to them is too low to begin with, or that they're usually constrained to formats that will require wasteful padding/dequantizing when read (at least for newer models) whereas a GPU just does that in local registers.
We don't ask that question. In reality everything is done in the cloud. Maybe they package some camera app that applies snapchat-like filters with NPUs, but that's about the extent of it.
Jokes aside: they really seem to do some things like live captions and translations. Pretty sure you could also do these things on the iGPU or CPU at a higher power draw.
No for sure, but afaik you get all of those features even if you don't have an NPU. And even if you do have one, it's unclear to me which one of them actually use the NPU for extra power or if they all just run on the CPU. Like the thing that is missing for me is "this is the thing that you can only do on a Copilot PC and it's not available otherwise".
Oh boy, instead of building an efficient index or optimizing the start menu or its built-in web browser, they're adding more power usage to make the computer randomly guess what I want returned since they still can't figure out how to return search results of what you actually typed.
It is another way Microsoft has tried to cater to OEMs as means to bring PC sales back to the glory exponential growth days, especially under the CoPilot+ PC branding, nowadays still siloed into Windows ARM.
In fairness NPUs can use less hardware resources than a general purpose discrete GPU, thus better for laptop workloads, however we all know that if a discrete GPU is available, there is not a technical reason for not using it, assuming enough local memory is available.
Ah, and NPUs are yet another thing that GNU/Linux folks would have to reverse engineer as well, as on Windows/Android/Apple OSes they are exposed via OS APIs, and there is yet no industry standard for them.
Thats the whole problem. No consistency. Some configurations work, others not - eventhough they should be way more capable.
That's not even limited to linux or gaming. A few weeks ago i tried to apply the latest Windows update to my 2018 lenovo thinkpad. It complained about insufficient space (had 20GB free). I then used a usb as swap (required by windows) and tried to install the update. Gave up after 1 hour without progress...
Hardware+OS really seems unfixable in some cases. I'm 100% getting a macbook next time. At least with Apple I can schedule a support appointment.
For gaming macOS does not seem a great choice. I have friends with macOS and, at least on Steam, there are very few games running on that platform.
Additionally when I was using macOS for work, I had also some unexpected things if I wanted to use anything a bit more special (think packages installed using homebrew, compiling a thing from source, etc.).
So for me the options are: either use a locked device where you can't do anything other than what the designers thought of and if you are lucky it will be good OR use something where you have complete freedom and take the responsibility to tweak when things dont'work. MacOS tries to be the first option (but in my opinion does not succeed as much as it claims to), Linux is the second option (but it is harder than it could be in many cases) and Windows tries to do both (and is worse than the two other alternatives)
For me it was about 8 years ago. Back then TF was already bloated but had two weaknesses. Their bet on static compute graphs made writing code verbose and debugging difficult.
The few people I know back then used keras instead. I switched to PyTorch for my next project which was more "batteries included".
Cheap labor. It doesn't take that much to train someone to be somewhat useful, in mmany cases. The main educators are universities and trade schools. Not companies.
And if they want more loyalty the can always provide more incentives for juniors to stay longer.
At least in my bubble it's astonishing how it's almost never worth it to stay at a company. You'd likely get overlooked for promotions and salary rises are almost insultingly low.
It would be nice if this model would be good enough to update their typscript sdk (+agents library) to use, or at least support, zod v4 - they still use v3.
Had to spend quite a long time to figure out a dependency error...
Kudos to bun for investing in a promising technology.
Does the Zig Foundation have a policy against corporate sponsors?
Otherwise the lack of sponsoring from the "big players" seems rather shocking. You'd think that zig has a decent chance in helping MS/Meta/Google/etc. somewhere along the way.
Since it's not a 1.0, it seems at face value it's be difficult for a "big player" to use it in production. As far as I understand, breaking changes are expected.
Yeah, makes sense, 1.0 is probably a critical point for a project like this, where from it, "big players" start trusting its business to the lang and therefore having a high interest on funding.
But it's kind of a chicken and egg problem: they need more money to keep doing its great work and thrive to reach 1.0 but good money comes from 1.0 and beyond.
Do you think absolutely all content should be allowed to be accessible?
If you wouldn't allow child porn (which 4chan deletes/doesn't allow), where exactly do you draw the line between blocking sites with cp, and allowing sites like 4chan which host porn without consent (voyeur/spy/revenge)?
There's a difference between prosecuting a crime, and restricting people to prevent it from even happening. Both have a place but only the former retains your liberty.
Yes, thats the problem. Prosecuting crimes on the internet is near impossible due to the restrictions and often anonymity. Thats why we rely on platform providers to help us, the public.
Facebook, Youtube and others put in effort to take down illegal content.
4chan only does the bare minimum such that they don't gain too much relevancy in the public eye.
UK or other countries may decide that 4chan doesn't to enough and ban it because of the help of 4chan in faciliating the spreading of illegal content.
So again, where is the difference between 4chan which hosts/spread illegal content and other sites where we're fine with banning them?
> Prosecuting crimes on the internet is near impossible due to the restrictions and often anonymity.
The US does it quite well. The FBI has a near endless number of court cases where they subpoena ISPs and content hosting platforms to de-anon and gather evidence to build cases. My biggest concern about the movement of crime from the streets to "cyberspace" is that almost all Internet crime is considered Federal (across state lines), thus carries much harsher penalties that state-only crimes.
There needs to be a greater distinction between models used for human chat, programming agents, and software-integration - where at least we benefitted from gemini flash models.
reply