dabockster

Jevons paradox - https://en.wikipedia.org/wiki/Jevons_paradox

It's the same problem with OpenRouter's free tiers for a long time. If something is truly $0 and widely available, people will absolutely bleed it dry.

This isn't true. I've gotten Claude and its forks to do things the same way as Copilot.

> so is probably only focused at local web design development vibe coding

Quick, someone throw the Linux kernel source at it and report back! xD

> We get on Zooms to talk. Even with the person 1 cubicle over.

Who normalized this?!!

> It is a vs code fork.

With vendor lock-in to Google's AI ecosystem, likely scraping/training on all of your code (regardless of whatever their ToS/EULA says), and being blocked from using the main VS Code extensions library.

> I don't know what IntelliJ is.

It started as a modernized Eclipse competitor (the Java IDE) but they've built a bunch of other IDEs based on it. Idk if it still runs on Java or not, but it had potential last I used it about a decade ago. But running GUI apps on the JVM isn't the best for 1000 reasons, so I hope they've moved off it.

US government bonds

> The failure mode of self-hosting is that your site gets hugged to death

Learn 2 load balance

The free credits... what a WILD time! Just show up to a hackathon booth, ask nicely, and you'd get months/years worth of "startup level" credits. Nothing super powerful - basically the equivalent of a few quad core boxes in a broom closet. But still for "free".

> At several startups, getting a simple €20–50/month Hetzner server meant rounds with leadership and a little dance with another department to hand over a credit card.

That's not a startup if you can't go straight to the founder and get a definite yes/no answer in a few minutes.

> But maybe things have changed in ways that meaningfully changes the calculus?

I'd argue that Docker has done that in a LOT of ways. The huge draw to AWS, from what I recall with my own experiences, was that it was cheaper than on-prem VMware licenses and hardware. So instead of virtualizing on proprietary hypervisors, firms outsourced their various technical and legal responsibilities to AWS. Now that Docker is more mature, largely open source, way less resource intensive, and can run on almost any compute hardware made in the last 15 years (or longer), the cost/benefit analysis starts to favor moving off AWS.

Also AWS used to give out free credits like free candy. I bet most of this is vendor lock in and a lot of institutional brain drain.

Apples and oranges, tbh.

You're literally playing into what the author is criticizing.

Depends on what VMware product you’re using. I’m still stuck on Workstation Pro because Hyper-V takes over my computer and slows down anything virtualized that isn’t Windows.

For servers, there’s things like Proxmox that are super versatile and can be deployed to a 1U server in a few hours. And if whatever you’re virtualizing exists mostly in the application/user layer and doesn’t touch lower level APIs (eg raw TCP), Docker/Portainer is even better (no Kubernates required!).

Or I just metaprompt a new chat if the one I’m in starts hallucinating.

From what I understand, only on the server level (just like the West). Desktops are still largely Windows or Mac.

And even then, the car is at a dealer and not your garage or some random shop. So it can potentially make the OEM liable if the update goes sideways.

I think that’s the big intent here. Make the operating system’s parental controls more all encompassing.

> So, an IT director might have to certify that all sysadmins and developers are over 18?

From what I understand, CA wants some kind of age field at the operating system/browser level that can be set once by a parent and trusted to work on every website out there - instead of the current system of needing to manually set an age on every individual website.

No it’s stupidity incarnate. Or these people have ADHD and are having it abused for corporate profits. Either way these people are going to be empty walking husks in under a year, likely for the rest of their lives.

Wasn’t there some kind of research that said something along the lines of human productivity dropping like a stone after 30-35 hours per week of work?

> Basically the Linux world was actively designed to apps difficult to distribute.

It has "too many experts", meaning that everyone has too much decision making power to force their own tiny variations into existing tools. So you end up needing 5+ different Python versions spread all over the file system just to run basic programs.

The big selling points for me were what you said about simplifying deployments, but also the fact that a container uses significantly less resource overhead than a full blown virtual machine. Containers really only work if your code works in user space and doesn't need anything super low level (eg TCP network stack), but as long as you stay in user space it's amazing.

> There is no need to invent new regulations.

The current regulations are written for a time where cars didn't have rolling computers in them. And even then, the regulations don't account for Tesla-style linked systems. So I say we do need new regulations.

Software updates for cars should be rare, with the infotainment systems physically isolated from the rest of the car (aside from constant/switched power). And if it needs updates for whatever reason (eg a recall), they should be done at a dealership - not over the air.

Bricking a car with an over-the-air software update is stupid and unacceptable behavior. Stuff like this is why I'm actually kind of happy to be driving a 20 year old car still.

The title is extremely misleading - you have to rent time on an H100 cluster to get it to work. It is not on-device, and thus not truly $100.

I was really excited, too, until I looked through the readme files and the code.

Discord's limitation is scaling as well, to be honest. It's incredibly hard to follow a server full of tens of thousands of people. Just because something can scale in a technical sense doesn't mean it will scale in a human one.

> although time to first token is... not great

The token cache works on CUDA too. So yeah, the initial loadout sucks, but almost everything from then on is solid.

But both Windows and Mac OS are developed largely in the US.

Could Trump unintentionally bring about the year of the Linux desktop?

The platform is just a search tool and extended processing modes. That is easily replicated elsewhere.

The only moat they have is the fact that you still need a GPU of some kind to reliably run even a tiny LLM. But the gap between what absolutely needs a server farm and what can be ran on a store bought gaming computer is quickly closing. You can already run mixture of experts models on gaming rigs with a high degree of usability compared to just one year ago. And that tech continues to be pushed further and further. It's only a matter of time until we see ChatGPT levels of access running on a quad core laptop totally offline. And once that happens, all such a system would need is the correct tooling on top of the AI model "brain" to make it usable.

And beyond that, what if you could have an AI model on an ASIC-style add-in card? Where's their moat then?

Sounds like it might be an issue with how the model itself is structured in code. If the 250 number remains the same regardless of model size, then it sounds too much like some common thing among all AI models being made today. GGML? PyTorch? Transformers? I think the issue lies in that area.