pca006132

Mesh is also a boundary representation. I think you meam NURBS.

LLM is just a tool. How the tool is used is also an important question. People vibe code these days, sometimes without proper review, but do you want them to vibe code a nuclear reactor controller without reviewing the code?

In principle we can just let anyone use LLM for medical advice provided that they should know LLMs are not reliable. But LLMs are engineered to sound reliable, and people often just believe its output. And cases showed that this can have severe consequences...

but how can you prevent the user from modifying the kernel?

By vendoring the code in, in this case I mean copying the related code into the project. You don't review everything. It is a bad way to deal with dependencies, but it feels similar to how people are using LLMs now for utility functions.

The problem is that it is natural to have code that is unreachable. Maybe you are trying to defend against potential cases that may be there in the future (e.g., things that are yet implemented), or algorithms written in a general way but are only used in a specific way. 100% test coverage requires removing these, and can hurt future development.

But this doesn't solve dependency hell. If the functionalities were loosely coupled, you can already vendor the code in and manually review them. If they are not, say it is a db, you still have to depend on that?

Or maybe you can use AI to vendor dependencies, review existing dependencies and updates. Never tried that, maybe that is better than the current approach, which is just trusting the upstream most of the time until something breaks.

Question: How many LoC do you let the AI write for each iteration? And do you review that? It sounds like you are letting it run off leash.

> as tasks that junior developers might perform don't match your skills, and are thus boring.

Yeah this sounds interesting, and matches my experience a bit. I was trying out AI for the Christmas cuz people I know are talking about it. I asked it to implement something (refactoring for better performance) that I think should be simple, it did that and looks amazing, all tests passed too! When I look into the implementation, AI got the shape right, but the internals were more complicated than needed and were wrong. Nonetheless it got me started into fixing things, and it got fixed quite quickly.

The performance of the model in this case is not great, perhaps it is also because I am new to this and don't know how to prompt it properly. But at least it is interesting.

> The idea of bitwise reproducibility for floating point computations is completely laughable in any part of the DL landscape. Meanwhile in just about every other area that uses fp computation it's been the defacto standard for decades.

It is quite annoying when you do parallelization, and idk if that many people cared about bitwise reproducibility, especially when it requires compromising a bit of performance.

Wonder if someone used effect handlers for error logging. Sounds like a natural and modular way of handling this problem.

It has many language bindings, including python and js. Though the js backend is not parallel because it uses wasm, and we had problem with mimalloc memory usage with pthread enabled.

At least 1 would not be enough. So how many branches are enough? And what about people with less money and time available?

But this is not related. You still have to pay the APC.

I thought this means for category theory people

anyway, quite cute :)

I remember people saying that chromium is better at sandboxing than firefox, so more secure.