I do quite like the stability of Cinnamon/Debian, and I think this problem is solvable (even if I have to solve it myself). I generally do not want to spend a lot of time futzing around with my desktop environment, but this is one thing I need to have.

I saw that and tried it pretty early on. That just moves the screen, it doesn’t fill the quadrant.

Updated to be specific, I’m using Cinnamon. Muffin is the builtin tiling window manager for Cinnamon and it does exactly what you’re describing. The problem is that it moves tiles, it doesn’t absolutely position them. You have to keep moving tiles around to get them where you want them, Rectangle just has hotkeys to immediately place and resize to fit the active window for each quadrant that it supports:

• ctrl+cmd+left: top left quadrant
• ctrl+cmd+right: top left quadrant
• shift+ctrl+cmd+left: bottom left quadrant
• shift+ctrl+cmd+right: bottom left quadrant
• alt+cmd+left: left half
• alt+cmd+right: right half
• alt+cmd+up: top half
• alt+cmd+left: bottom half
• alt+cmd+f: full screen

It’s hard to express how natural that feels after using it for a bit, and I’m still using a Macbook for work so the muscle memory is not going away.

I didn’t say it wasn’t amazing nor that it couldn’t be a component in a larger solution but I don’t think LLMs work like our brains and I think the current trend of more tokens/parameters/training LLMs is a dead-end. They’re simulating the language area of human brains, sure, but there’s no reasoning or understanding in an LLM.

In most cases, the responses from well-trained models are great, but you can pretty easily see the cracks when you spend extended time with them on a topic. You’ll start to get oddly inconsistent answers the longer the conversation goes and the more branches you take. The best fit line (it’s a crude metaphor, but I don’t think it’s wrong) starts fitting less and less well until the conversation completely falls apart. That’s generally called “hallucination” but I’m not a fan of that because it implies a lot about the model that isn’t really true. Y

You may have already read this, but if you haven’t: Steven Wolfram wrote a great overview of how GPT works that isn’t too technical. There’s also a great sci-fi novel from 2006 called Blindsight that explores the way facsimiles of intelligence can be had without consciousness or even understanding and I’ve found it to be a really interesting way to think about LLMs.

It’s possible to build a really good Chinese room that can pass the Turing test, and I think LLMs are exactly that. More tokens/parameters/training aren’t going to change that, they’ll just make them better Chinese rooms.

Maybe this comment will age poorly, but I think AGI is a long way off. LLMs are a dead-end, IMO. They are easy to improve with the tech we have today and they can be very useful, so there’s a ton of hype around them. They’re also easy to build tools around, so everyone in tech is trying to get their piece of AI now.

However, LLMs are chat interfaces to searching a large dataset, and that’s about it. Even the image generators are doing this, the dataset just happens to be visual. All of the results you get from a prompt are just queries into that data, even when you get a result that makes it seem intelligent. The model is finding a best-fit response based on billions of parameters, like a hyperdimensional regression analysis. In other words, it’s pattern-matching.

A lot of people will say that’s intelligence, but it’s different; the LLM isn’t capable of understanding anything new, it can only generate a response from something in its training set. More parameters, better training, and larger context windows just refine the search results, they don’t make the LLM smarter.

AGI needs something new, we aren’t going to get there with any of the approaches used today. RemindMe! 5 years to see if this aged like wine or milk.

Hyperfixating on producing performant code by using Rust (when you code in a very particular way) makes applications worse. Good API and system design are a lot easier when you aren’t constantly having to think about memory allocations and reference counting. Rust puts that dead-center of the developer experience with pointers/ownership/Arcs/Mutexes/etc and for most webapps it just doesn’t matter how memory is allocated. It’s cognitive load for no reason.

The actual code running for the majority of webapps (including Lemmy) is not that complicated, you’re just applying some business logic and doing CRUD operations with datastores. It’s a lot more important to consider how your app interacts with your dependencies than how to get your business logic to be hyper-efficient. Your code is going to be waiting on network I/O and DB operations most of the time anyway.

Hindsight is 20/20 and I’m not faulting anyone for not thinking through a personal project, but I don’t think Rust did Lemmy any favors. At the end of the day, it doesn’t matter how performant your code is if you make bad design and dependency choices. Rust makes it harder to see these bad choices because you have to spend so much time in the weeds.

To be clear, I’m not shitting on Rust. I’ve used it for a few projects and great for apps where processing performance is important. It’s just not a good choice for most webapps, you’d be far better off in a higher-level language.

I wouldn’t shortchange how much making the barrier to entry lower can help. You have to fight Rust a lot to build anything complex, and that can have a chilling effect on contributions. This is not a dig at Rust; it has to force you to build things in a particular way because it has to guarantee memory safety at compile time. That isn’t to say that Rust’s approach is the only way to be sure your code is safe, mind you, just that Rust’s insistence on memory safety at compile time is constraining.

To be frank, this isn’t necessary most of the time, and Rust will force you to spend ages worrying about problems that may not apply to your project. Java gets a bad rap but it’s second only to Python in ease-of-use. When you’re working on an API-driven webapp, you really don’t need Rust’s efficiency as much as you need a well-defined architecture that people can easily contribute to.

I doubt it’ll magically fix everything on its own, but a combo of good contribution policies and a more approachable codebase might.

Why do you think ventilators made people worse? They only put people on ventilators when their O2 stats dropped so low they were going to die of oxygen deprivation.

Part of the reason these rules are similar is because AI-generated images look very dreamlike. The objects in the image are synthesized from a large corpus of real images. The synthesis is usually imperfect, but close enough that human brains can recognize it as the type of object that was intended from the prompt.

Mythical creatures are imaginary, and the descriptions obviously come from human brains rather than real life. If anyone “saw” a mythical creature, it would have been the brain’s best approximation of a shape the person was expecting to see. But, just like a dream, it wouldn’t be quite right. The brain would be filling in the gaps rather than correctly interpreting something in real life.

I can hear René Auberjonois in that line