Aozora7

We'll probably get there, once frontier labs RL the fuck out of AI agents sitting in meetings and interpreting incoherent ramblings of business majors into actionable requirements. That's the biggest value human devs currently provide.

Yes, AI for now has significant code quality issues, but that's mostly because it lacks agency to take care of code quality unless you explicitly tell it to. It is good at refactoring its own messes when you even vaguely ask for it. So while something like Astra still needs supervision to produce decent code, I expect that in a year or two it will be unnecessary.

I have seen a decent amount of clearly AI generated software that's useful, would likely not have been made without AI, and is better for for being AI generated. With AI, people make improvements that they would have never have bothered spending time to do manually. If anything, having uncharacteristically good "production value" in a clearly hobby project is how I typically spot AI use in projects that aren't obvious vibe coded slop.

The code itself may or may not be spaghetti. Not that I care as a user. User experience and code quality had never a particularly strong correlation even before AI.

No. Before AI, I have never wrote any code outside of work because nothing I could do in a reasonable time would be worth the effort.

With AI, if there's software I wish existed, I can make it happen without making it an unpaid second job.

Isn't the main issue that the audio you're compressing over bluetooth, like spotify streams, is already compressed? As far as I know, compressing already compressed audio compounds artifacts in the same way re-saving a JPEG does.

I switched to Firefox when chrome announced manifest v3, and I'm still happy with it. Not only I get a browser with a real adblock, now that I got used to multi-account containers, I can't go back to chrome since it has no good equivalent. And I'm also using it on mobile since it's one of the few mobile browsers with extensions too.

Noise reduction benefit is marginal. Most of the noise comes from tires rather than the engine. At 30 km/h tires are louder than the engine, so an EV driving at 50 km/h, which is the typical city speed limit, isn't any quieter than an ICE car.

In this particular case, it would be harder to produce a fake signature from a non-existent person if the signatures in questions were digital and backed by a national identity number. The kind of that I'm submitting when signing documents at work.

I'll have you know I started picking salads instead of fries in combo meals the moment they became available.

Not that I'm a frequent customer, either before or after.

When you renew your passport, the new one will have a new serial. Not matching a passport serial doesn't mean anything. Not matching a national identity number does.

Very few countries require you to carry an ID. Many countries require you to own one and keep it up to date, but don't care whether you carry it outside.

Also, I find the point kind of moot for the US where a driving license is essentially an ID card, you have to carry it to drive, and you have to drive to go anywhere in most places.

A government won't give up on malicious acts just because it lacks accurate census. The United States does a lot more surveillance and oppression than EU countries that have national ID and mandatory ID documents.

The lack of a comprehensive population database only seems to hinder the actual useful civilian bureaucracy, not law enforcement, intelligence services, or ICE. The latter just grab anyone who looks brown enough.

Identity numbers being abused is a consequence of them being misused. And as far as I can tell, this happens more often with american SSNs than anywhere with actual national identity numbers.

All that these numbers do is uniquely identify people. They aren't private. To actually prove your identity, you present your ID, passport, or a digital signature tied to that identity number.

This is one of the obvious consequences of not having a national identity number assigned to everyone at birth, which exist in most developed countries, except for anglosphere, for some reason.

Harnesses exist to give models tools to do work beyond generating text. People install Claude Code and Codex to have models work inside their repositories and run the code or tests themselves instead of the user having to copypaste code between their IDE and a chat interface. The fact that there's some security built into the harnesses is just a practical consideration, not its primary function.

Consider what someone who doesn't use AI would take away reading that quote from the article, and what harnesses were actually made for.

I personally find that I hate any activity in a context where I'm paid for it.

I used GLM 5.0/5.1/5.2 for some projects, and for me, the area in which they lag behind frontier models the most are user interfaces. They get really close to Opus when it comes to pure algorithms, but when I need something like web application or a mobile app that looks and works well, they are very noticeably worse than even Sonnet.

There is, for example, OpenCode Go subscription, which for $10 a month gives you a decently generous quota of GLM-5.2, among other models.

And z.ai themselves also have subscriptions.

Author is from Latvia (and so am I). You do actually get carded for energy drinks if you look under 30.

However, more relevant to the post, is that when you're ordering groceries online, you need to verify your age at checkout if you're buying stuff like alcohol (or energy drinks). It's trivial, and for a lot of people it uses the same authentication service that they already use to access their bank.

With the release of GLM-5, I would say that they are pretty much almost as good. Basically 90% as good as Opus 4.6 on most tasks for 20% of inference cost, and open weights.

There already are LLMs with open weights that are better at code than state of the art closed source models from a year ago. For now, you most people may have to rent the hardware to run those models, since it's too expensive for most people to own something that can run inference on one trillion parameters, but I wouldn't consider LLMs to be controlled by "evil software corporations" at this point.

>Perhaps with time we'll be able to run local ones that are 'good enough', but we're not there yet.

Right now, we can get local models that you can run on consumer hardware, that match capabilities of state of the art models from two years ago. The improvements to model architecture may or may not maintain the same pace in the future, but we will get a local equivalent to Opus 4.6 or whatever other benchmark of "good enough" you have, in the foreseeable future.

I'm not paying for an AI subscription to do my job in the same way I don't pay for the IDE I use. My employer does.

>Because they were trained for OTHER work!

Not by choice though.

Vast majority of people who get CS degrees don't want to be computer scientists. They want to do programming work, so they get a CS degree because that's the degree employers require. They don't even know that said degree won't focus on what they actually want to learn.

And it's not like they have better choices. There are no programming trade schools. Bootcamps seem to have the right idea in principle, but implementations are often questionable, and most employers won't consider bootcamp graduates.

I'm using 4 monitors, two 33" and two vertical 22".

It has nothing to do with knowing how to do shortcuts, but having to do them at all. There are plenty of instances where you may only need to provide input to one window at a time, but see other things, like documentation, output, logs, or a browser window of the page you are editing. Having multiple monitors allows me to greatly reduce the need to switch between windows to reference something or see results of my inputs.

I personally find that if I'm forced to alt-tab between my IDE and documentation, like when I'm working on a laptop, it's incredibly distracting, breaks my flow, and slows down my work significantly.

I have the opposite experience, the CLI users I see pretty much never use anything besides commit, push and pull, while GUI users tend to use more advanced commands since they can actually see what they do.

I find git reasonably understandable as long as you are using a GUI. When you have a GUI, every action you take provides you visual feedback, so it's way easier to understand the purposes of merges, rebeases, stashes, and resets. My understanding of git workflow greatly improved thanks to it.

It was counterproductive for me when recursion was taught with Fibonacci and factorials. They were literally the opposite of easy to understand for me, I had absolutely no idea why those examples actually worked, so I just gave up and treated them like magic.

I think recursion should be taught with examples that would be easier for a student to write with recursion than without it. For example, listing all files in a directory.

You don't have to hunt for rare examples like this, most multiplayer FPS games use client-side hit detection nowadays.

What I personally see in academic setting is that when recursion is taught, it's taught using examples where you would not want to use recursion.

It's easy to understand why recursion is useful when you are tasked with traversing a file system. It's not when the first example you are given is calculating Fibonacci sequence.