The AI technofacists building these systems have explicitly said they’ve hit a wall. They’re having to invest in their own power plants just to run these models. They have scores of racks of GPUs, so they’re dependent upon the silicon market. AI isn’t becoming “ever more capable,” it’s merely pushing the limits of what they have left.

While I agree that this paper sounds like a freshman thesis, I think you’re betraying your own lack of knowledge here.

Because no, they havent said they’ve hit a wall, and while there are reasons to be skeptical of the brute force scaling approach that a lot of companies are taking, those companies are doing that because they have massive amounts of capital and scaling is an easy way to spend capital to improve the results of your model while your researchers figure out how to make better models, leaving you in a better market position when the next breakthrough or advancement happens.

The reasoning models of today like o1 and Claude 3.7 are substantially more capable than the faster models that predate them, and while you can make an argument that the resource / speed trade off isn’t worth it, they’re also the very first generation of models that are trying to integrate LLMs into a more logical reasoning framework.

This is on top of the broader usage of AI that is rapidly becoming more capable. The fuzzy pattern matching techniques that LLMs use have literally already revolutionized fields like Protein Structural Analysis, all the result of a single targeted DeepMind project.

The techniques behind AI allow computers to solve whole new classes of problems that werent possible before, dismissing that is just putting your head in the sand.

And yes companies are still dependent on silicon and energy, which is why they’re vertically integrating and starting to try and produce that on their own. That’s not a sign that they see AI as a waste of time.

And to make sure that everyone is portrayed by glen powell

You could but you’d be drawing a false equivalency.

Do you have any resources by any chance that explain the difference well?

I work in high level software, so understand the benefit of doing things at ide time vs compile time vs runtime, and I’ve coded in assembly back in the day and understand instruction sets at a very rough level, but I’m not really familiar with specifically what differentiates RISC / ARM / x64, or why RISC’s reductions would be good / bad / what trade-offs come with them.

Fair point, I misspoke / don’t actually think that it’s wild, I was being dramatic.

Have you tried using VSCode / VSCodium? I’ve tried using a VIM based workflow and found myself missing many graphical dev features in VSCode.

And sure, there’s nothing wild about continuing to use a process that works for you, but it is a little wild to insist that your process is the best and other people should learn it, if you also know that it has inherent limitations that alternatives don’t.

Fair enough!

K, now give me the longest edge and it’s displacement relative to the x axis. Then rotate the shape until that edge is roughly 33 degrees off the z axis.

Oh wow, look, suddenly it may be helpful to have a way other than text to draw and visualize things.

My worldview is that it’s wild to choose a dev tool chain incapable of drawing basic 2d shapes, when you have ones available that can do anything.

But it’s literally doing that in your image. When a horizontal and vertical line cross the horizontal line breaks.

Yes, as an intentional graphical choice to illustrate the crossing of two paths.

In lazyvim a vertical line, with no crossings, is still broken, as it is two pipes separated by the line space height.

Oh, did you mean the points that represent actual commits? You’re arguing it’s trash because there’s no line between two adjacent commits? Really?

No, I’m saying it’s trash because it CANNOT do something basic like drawing a continuous vertical line, because it is hamstrung by using the interface of a typewriter. A git branch is just one readily available example of a situation where something extremely basic like drawing a continuous line would make sense.

You’ve brought it up multiple times now so I think it’s time you also source that claim. Cmon, source the claim where the code editor with better visual fidelity increases productivity.

I can’t cite internal market research that is under NDA. I can point you to basic courses on design and UX, point you to information on concepts like cognitive overload, and point out to you the multiple trillion dollar software companies that got to where they are entirely through paying attention to little UX details that backend nerds previously claimed didn’t matter and were user skill issues.

Yes, terminal can’t do everything, but I don’t think anyone is using VS code to look at a cube either. Actually, I’m not even sure if there is a VS code extension that draws cubes? So you wouldn’t use VS code for that either.

Bruh, why would you even try and talk out of your ass like this? I am literally using jsCad and VsCode to do my personal 3d printing modelling, and I literally got my start programming using first VS, then VSCode, to build 3d modelling software for Autodesk. Not sure if you’re aware of this but modern websites have this little thing called WebGL that lets them display these little things called jraphics.

Again, VsCode can do everything VIM can do, but not vice versa.

and if I do not want the GUI part, how come it surprises you that I do not use that superset?

Go ahead and represent an arbitrary 3d shape using the command line, suddenly you may realize that a typewriter’s interface isn’t the fastest for accomplishing every programming task.

Regardless, you can be happy with a limited subset of functionality and trying to cram every interaction into text, that’s not an argument that that way is better or that a new dev should go that route, just that you can get by using that method.

I said continuous vertical lines and literally posted a screenshot of it not being able to do it.

It’s functionally the same visual representation of data so you’re literally arguing over it not looking like you want it to look.

No, it’s not. The human brain does not process dashed lines as easily as it does continuous lines. A whole bunch of dashed lines are objectively harder to follow than continuous ones.

You can think that’s not important, but the literal decades of UX research and attention to fine grained user interaction, can prove that you’re just flat out wrong.

You look at the above and think they’re the same, but they’re fundamentally not. Literally just go ahead and try and visualize a basuc cube with this base point and dimensions through a CLI and watch that wow, maybe a fucking typewriter interface isn’t the best for absolutely everything:

Cube([0.37, -300, 45], [37,-98,-100])

No, the conclusion I’ve been saying is that CLI developers are smart people who have spent a long time memorizing commands to get fast at things that can be done quickly and intuitively through basic 2d graphical interfaces.

They’re now either in a situation where the gains from learning the new process aren’t going to outweigh the costs (though still doesn’t mean anyone else should follow their path), or they would, but they’re just stuck in their ways because of sunk cost fallacy.

No, I’m not. I’m just pointing out how lazygit is still limited by being a line by line, text based, CLI interface, and thus cannot draw a continuous vertical line, even if drawing a continuous vertical line would make sense in that situation:

Ok, cool beans bro, try and write 3d modelling software with just a command line interface and you’ll quickly see how a typewriter’s format for displaying text isn’t the fastest for every programming task.

But the VSCode plugin ecosystem still lacks some features available in the Vim ecosystem, and (fl just for example)

Isn’t that basically the same as Command Shift P and / or the search feature?

At the end of the day, the biggest difference is speed. Even very brief unexpected delays can break my concentration. While VSCode is no slacker, it still has some delays, probably mainly because it’s still JavaScript under the hood.

Once there’s a GoLang, Rust or C port of VSCode, I may well switch permanently.

I can 100% understand how big of a deal speed delays can be, but at the same time, not to probe too hard, but what are you experiencing delays in? In all honesty waiting for ohmyzsh to start, or waiting for a git pull to run, takes far longer than any task I can think of in VSCode. Files open faster than notepad, the file browser is fast, the shortcuts and commands are fast, I honestly haven’t experienced any slow downs with it anywhere, and I’ve used it with monorepos that are TB in size.

Literally not, since I’m advocating for a superset of what they are.

I use command line tooling perfectly happy within VSCode, they don’t use graphical tooling within VIM.

I’m literally just advocating for a toolset that lets you use graphics or a cli, depending on what makes most sense for the task at hand, they’re advocating to only use the cli.

Sure, if the only criterion you are trying to fulfill is “have as many options and different ways to complete the task at hand as possible,”

Except that’s not what I’m saying.

I’m saying it’s important to have the right tool available for the job.

If you limit yourself to VIM and command line interfaces, it will mot matter if a GUI is the right tool, it’s not in your tool chain, you can’t use it.

i.e. I don’t use VSCode because it provides me with multiple ways of viewing git’s branching history, I use it because it provides me with the better way of doing so. And when the better way of doing something involves using the command line, it lets me do that too.

People insisting on using the command line for everything is like a carpenter that only buys a circular saw and refuse to buy any other saws. Like yeah, you can do almost any cut with a circular saw, and it’s not a bad place to start, but theres a reason professional carpenters who need to do repeated cuts quickly, accurately, and in a way that is teachable to others, don’t limit themselves to a single type of tool for every scenario.

Getting an automatic terminal window when you start up vs code is no different having two panes in tmux, one for VIM and once for terminal.

Yes it is, and I honestly cannot fathom how you cannot seem to comprehend the difference between text, and an actual pleasant to use and look at graphical interface.

Lazygit looks exactly as trash as the OOTB command line git. How do you not understand that the human brain processes a smooth connected line more easily than a pseudo line broken up by the line space height, made out of pipes and slashes? This is like product design and UX 101.

Again, VSCode does everything VIM does. Not vice versa, one is a superset of the other.

GUI has objectively way more visual noise

Nope. You can open up VSCode and just have it open to a terminal window if you want.

A GUI + Terminal gives you more options than just a terminal. It’s not complicated and it’s not arguable, one is a superset of the other.