anaisbetts

5,292 karmajoined 18 anni fa

Client Infrastructure at Anysphere/Cursor

comments

4 giorni fadiscuss

The extremely charitable answer is that this is AMD's chance to control the built-in software, so they can ship a bunch of built-in AI stuff and a controlled environment similar to what NVIDIA does with their DGX OS, rather than e.g. Framework which is going to give you a stock Linux distro.

(I agree that you almost certainly don't need this)

14 giorni fadiscuss

I mean just like GGUFs aren't technically necessary yet are _way_ more convenient than using Safetensors and configuring the default Jinja prompt by-hand, it makes sense to bundle the draft model too. For all intents and purposes, the only people who will train a draft model are the people who train the original model

17 giorni fadiscuss

The exit tax doesn't apply to "gains", it applies to the "value of your company" which is calculated in a way that often means you will owe thousands or even millions in money you don't have, and at no time had.

21 giorni fadiscuss

The vast majority of children cannot and do not buy and pay for VPNs and use them to evade anything. The second that governments said "We intend to ban VPNs" they gave away the game as to what their actual intent was, and it wasn't "for the children"

21 giorni fadiscuss

This article is likely AI slop, none of this is "news" and I'm pretty sure that this is just factually Not Accurate. Windows 11 is not shipping a new media player, it has the same one as it has since Win10. Codecs have like HEVC/H.265 always cost a very small amount of money, in order to pay patent owners. Again, it has done this for years now.

25 giorni fadiscuss

If you're hitting this you have a bug, this is not related to the model. Either your harness is editing the messages between turns incorrectly (i.e. it is not append-only), or sometimes this is because of llama.cpp bugs, but bet on the former. Setting up something like Tailscale's Aperture will let you capture the requests and then you can diff them.

mese scorsodiscuss

No headless RDP solution is pretty unforgiveable in 2026 - right now I use Nomachine and indeed it requires X11. Why wouldn't Wayland have a solution for this? It is not a niche usecase. It is like, a headline, completely mainstream use-case.

mese scorsodiscuss

Sure but MediaTek isn't making the actual CPU, they're just making the unimportant parts of the chipset, who cares

mese scorsodiscuss

It's an ARM64 CPU with a nVidia GPU rather than garbage Adreno, not bad

2 mesi fadiscuss

Proton Mail not supporting filters for message bodies is brutal, I understand why they don't do it but that really lowers its usability for me. Bummer.

2 mesi fadiscuss

Honestly, for most open-source apps where you don't want to Deal with this, the answer is to just get it into Homebrew. They handle re-signing apps using the local computer's ad-hoc signature and for most apps this is by-far the easiest way to deal with Stupid Apple Bullshit.

2 mesi fadiscuss

I don't think this is true. A normal CS signing cert is sufficient for most commercial apps - you will get the SmartScreen warning for a few days but it will go away fairly quickly.

The important part is that SmartScreen reputation is URL-based, you need to make your initial download URL consistent. If you are constantly rewriting it (i.e. with a version #) it will break. It's ok if the original URL hits a 302 to the latest version.

3 mesi fadiscuss

This hasn't been my experience, ROCm is usually not only a bit slower for me (~32 t/s vs ~43 t/s on the main model I use), it is way less reliable; any upgrade in kernel version or AMD driver and suddenly everything is broken

3 mesi fadiscuss

Sure but on the flip side, skills not being in context means that for many harnesses, the model simply never finds them. Whether MCP or Skills are "better" depends extremely heavily on the context management functionality of your harness because if you use a relatively naive harness (i.e. one that implements MCP and Skills in a straightforward way), MCPs will generally be more effective, especially if your model is local-only (i.e. dumb), but at the cost of context.

5 mesi fadiscuss

Came in getting ready to hate on this article, but was actually pleasantly surprised, this is great. Good work OP.

5 mesi fadiscuss

That'll work great until your first customer from a CJK or RTL language writes in, "Hey, how come I can't type in your app?", or the blind user writes in "Hey how come your app is completely blank?" then you'll be right in the middle of the "Find Out" phase

These strategies are fine for toy apps but you cannot ship a production app to millions or even thousands of people without these basics.

6 mesi fadiscuss

A pervasive "Someone needs to do something!!!" attitude is why. Americans will forever wait for the school principal to come and get everyone into trouble

6 mesi fadiscuss

The code generator just recreates the C header files in C#, let's not be dramatic.

6 mesi fadiscuss

If you've come to Germany as a foreigner recently you might have it automatically, newer ID cards have it activated

6 mesi fadiscuss

The country that lived through pervasive mass state surveillance by secret police for 40 years is unsurprisingly quite cagey about digital centralization of records, even so many years later