vanyle

I don't think this is an emergent capability. I think Astra was trained on neural weights data as an attempt by OpenAI to get to recursive self improvement.

Something similar happen in early GPT models which had a lot of chess game in standard notation in their training data which made them very good at chess as long as your prompt had the right format.

We'll see if this attempt at RSI worked.

I wrote an automated testing system for my game engine to take screenshots and simulate input for automated testing and then I realized how good claude was at using it to make games that I wrote a guide about it in my engine documentation:

https://vectarineengine.com/guides/working-with-agents/

The short story about the digital animals ("The Lifecycle of Software Objects") is an interesting read in light of AI advancement. It is incredible to think that when this story was written, all concepts described were sci-fi, whereas current technology is more advanced than the AIs in the book.

This paper is a large amount of nothing. First, natural language is not as precise as lean, so you have multiple ways to translate a NL argument to Lean. As shown in Fig 1, the LLM did a decent job at translating the argument about roots in a succint way.

Moreover, the paper claims that the NL arguments of Navier-Stokes are stronger than the Lean ones. My understanding is that the translator LLM got lazy and wrote the minimal amount of code that satisfied the theorem without the extra stronger claims.

It is common in mathematical papers to say "And by the way, this actually proves [stronger claim]", but this is something an AI with a precise goal of performing a translation would never do, as it's goal is to translate the proof, not to quality mathematics.

I think this move will be part of a more global trend when other AI companies get publicly traded.

Because model providers have no moat, and the price of inference is quite large compared to just search, providing it for free makes no sense.

Right now, the goal of the free chatgpt and claude tiers is to accustom more people to AI, to create a new market, not to capture customers.

This strategy makes sense for start-ups with a lot of funding, but not for companies that need to be profitable. Apart from google, no large company is providing free inference to their large models in a consumer app (apart from meta maybe).

I've personally been using vite with turbo, pnpm, rolldown and oxlint, as separate tools and it is working very well.

I don't really see the value for me in having these be bundled together, but I can understand why it makes a monorepo setup simpler for a newcomer.

I think the issue comes from the practice of always running "git add ." from the cli. As a git gui user [1], I see exactly what files I stage, so this is a non-issue.

Also, the .gitignore file can document what files you are supposed to have in your project and where they come from

[1]: I'm currently a happy Fork user after having tried Sublime Merge, Git Kraken and lazygit.

This reads a bit like "Door locks do not survive contact with reality."

The point is not to prevent state-sponsored actors to produce fake media, but to add friction and avoid shady marketing agency and photographs from claiming their pictures are genuine.

vectarine https://github.com/vanyle/vectarine https://vectarineengine.com

It's a game engine with hot-reloading for everything to build and edit games while they are running

I find the concept of having multiple types of personalities that are to love in their own ways interesting. I reminds me of the love languages classification.

Centralized package repositories like the one provided by canonical have similar limitations to the Mac App Store, you need to get your app reviewed, you need to push updates to each platform where you distribute your app and in exchange you get visibility.

I'd argue that installing and updating apps on MacOS is simpler than on Linux distros because most apps have built-in auto-updates (or you can just drag the app to the applications folder) instead of having to rely on snap / apt / insert your package manager which may a lot of outdated and unmaintained packages and apps.

I tried raylib and I love it, but I need to build a lot from scratch (like most game frameworks). But I also really dislike engines as I prefer making games with code rather than a GUI. I am currently working on a framework/engine hybrid called vectarine [1] where I make my game with code while still enjoying niceties of an engine like hot reloading, integrated debugging, asset management etc.

[1] https://github.com/vanyle/vectarine

Having built crossplatform native app supporting MacOS app, I have never gotten notarization to work as I'm not using xcode a lot.

I'm curious, how much does it cost? Is it per build or a subscription? How do you make it work financially for an open-source project?

The git seems to only contain the build of the website with no source code.

The author is probably using git to push the content to the hosting server as an rsync alternative, but there does not seem to be much leaked information, apart from the url of the private repository.

For CLI tools, the program usually only executes for a few seconds and then returns. In this context, you don't even need arena allocators or any memory management at all. You can write the tool in C with malloc and let the OS free the memory at the end without worry.

You could argue with the reasoning that C feels more practical than Zig for real-world CLI tools.

The argument provided by the author feels a bit besides the point.

As a workaround, you can add a transparent border (border: 2px solid transparent) around the skewed element to have antialiasing (at least on chrome)

Workday good actually! Yes really nice