I just have small thing to add to this article - it mentions how the code contributed per engineer has increased as per Claude Mythos to 8x of baseline.
Now, I have encountered many times, when I asked AI to implement a function for me for which I was 100% sure a good implementation already existed in the form of an npm package, it had the tendency to go ahead and implement it on its own. Now, I usually trust battle tested implementations to be more robust, but if the AI does this (which I think is not an unique observation), you can easily balloon per engineer line generation (as can you with reduced oversight), so as always, these high level benchmarks are to be taken with a grain of salt.
Maybe Im nitpicking here but LLMs are quite literal. So when you tell it to "implement a function for me" it will necessarily write the whole thing. Changing the prompt to "find an existing implementation for this" would be more apt.
Imo it was the right call. Linux filesystems, file attributes are different from Windows, and the two never would've been completely compatible. The two big ones, are the execute bit, and the very fact that windows files represent the data on disk, while Linux ones are just a hardlink to the inode.
It is what it is. My workflow is that I rsync the relevant files from Windows to Linux, do the work, and rsync them back. It's clunky but works well enough.
I, who has to professionally support installs running Linux with Nvidia hardware, would personally say the situation is very far from ideal on Linux.
I dislike this 'my dad can beat yours' kinda competition when it's very clear Linux still has significant issues to resolve.
Proprietary for now, not sure how good the open source one is. But the proprietary has many quirks/bugs and limitations especially when i comes to things like Linux specific Vulkan extensions.
I don't know the full story, but afaik, AMD/Intel Mesa based drivers are fully open source and are built in a much more Linux-native way. But unfortunately hardware choice is out of my jurisdiction.
WSL2 operates through a file-level translation layer, not unlike NFS in Linux, which can be a huge performance problem for programs that manipulate lots of tiny files, like git or npm.
If I remember correctly, Zed's framework didn't set the goal of being able to draw arbitrary graphics/UI and by constraining that, it basically managed to represent everything with quads and distance fields in shaders, which reduced draw calls and GPU state management to a minimum.
Not anymore. I bought a brand-spanking new high-end Electrolux dishwasher and discovered it had a flimsier construction than the 20 year old one my mom has at home, that was an absolute bargain bin purchase.
From what I've seen of youtube teardowns by mechanics, Chinese cars often have the air of modernity and future tech about them, but under the hood, they like to use technical solutions that were abandoned by European manufacturers a decade ago or more, leading to the cars not being competitive in driving dynamics to what you would get from an euro manufacturer for perhaps slightly more money.
European cars are incredibly expensive these days and have been for a while compared to imports.
About 7 years ago I was looking for a new car and had only owned and drove VW cars. Naturally, I looked at VW, the Golf in particular. With all the bells and whistles, it came to about 35k EUR. A Civic was about 11K cheaper. Guess what I bought?
There was nothing in the Golf that could justify the extra cost. Nowadays is the same.
More fun driving dynamics aren't worth the extra twenty thousand dollars you need to buy the equivalent European car, let alone the ridiculous lifetime costs of ownership, such as maintenance.
I mean, seriously. I can buy a shitty Argentine-made base model used 5-year-old VW Amarok with 70K km on it for the same price as a brand new Chinese-made Changan Hunter (whether gas or REEV). The second one is also going to be far, far cheaper to fuel (if you choose the REEV), maintain and repair.
Let's not even talk EVs. Europeans exited the market before it even began. European brands only start to show up at around what, the 70-100K dollar bracket? Sure, the Porsche Taycan is cool and all, but I've never seen one. Meanwhile, BYDs, Deepals, MGs, and a million other random brands are only growing more and more common with each passing day.
I'm not American, but is talking to bartenders actually a thing? When I go to places like this I always either go with friends/family. There are some smaller pubs where I do know the bartender/pub owner, but it's more of a familiarity/community thing.
I've always assumed that a guy walking into a bar and discussing his intimate life with a person he just met is more of a Hollywood plot hook.
Vertical integration is a double edged sword. When your car is a decade old or more, and stuff starts breaking, the only vendor that makes parts for it is BYD - and its up to them if they bother selling to you, at what price, provided they still make these parts at that time.
Time will tell how cheap and easy will they be to maintain, which affects residual value at the end of the lease, which affects payments.
Nobody pays sticker price for new cars as a lump sum. If BYDs (or whatever car) are impossible to maintain, that means nobody will want them used, and residual value will be low, which means BYDs will cost more to the end user to own than a more expensive car by a different manufacturer.
Which seems to be the case for now in places like Germany, but we will have to wait and see how the situation develops and the second hand market builds up.
Additionally, vertically integrated companies don't get to tag along on innovations produced by upstream suppliers. Many EV companies will effortlessly have their range increase by a couple % each year, with battery costs falling, just because they can adopt the improvements of CATL/Panasonic/BYD etc.
The whole point of Hare is to be a part of a modern computing environment that offers most of the creature comforts of modern computing, yet is small enough to be understood by one person.
The Linux kernel ABI, while stable, is not simpler than libc and not portable at all. You can build the modern computing environment on mingw.
However, if they chose to target Linux–only to show how low–level Hare is, that's understandable as well —no shade.
To take it to its logical conclusion, they're saying libc will be a package you install on your Hare/Linux system for compatibility with obsolete systems, which to me implies willingness to work on ports to other platforms —you're not going to replace libc otherwise.
Now, I have encountered many times, when I asked AI to implement a function for me for which I was 100% sure a good implementation already existed in the form of an npm package, it had the tendency to go ahead and implement it on its own. Now, I usually trust battle tested implementations to be more robust, but if the AI does this (which I think is not an unique observation), you can easily balloon per engineer line generation (as can you with reduced oversight), so as always, these high level benchmarks are to be taken with a grain of salt.
reply