edflsafoiewq

I think this narrow view on determinism comes up a lot. Essentially any program can be made deterministic over single inputs by fixing all side-inputs. But there is determinism over classes of inputs too, eg. an algorithm given input X deterministically outputs X+1.

I think the wider view is usually what people mean when they talk about nondeterminism in LLMs. Maybe because computers are traditionally so deterministic the wide view is almost taken for granted by programmers.

Not if you're not giving feedback.

Presumably the same problems exist for avif/jxl?

Here some quick thumbnails for all of them: https://i.imgur.com/KXjiYxO.png

Use redlib. It's crazy how much better the interface is than both old and new reddit.

> which have been a nightmare on mastodon

Why?

Get a whole "for parts" model off ebay. The knockoff keyboards suck.

They use their own llama fork anyway, so that shouldn't matter.

I think the general idea is naive quantization falls apart below 4bpw but you can go lower with more sophisticated QAT-adjacent methods. Bonsai's quantization method is proprietary though.

If the s are small the usual geometric buffer growth mitigates that. Of course you can compute the final buffer size in this case, but often you have a bunch of dynamically-generated strings of different sizes.

Does Python actually do that? If the f call throws, you can still observe the (unchanged) binding of acc in reduce.

The binding for acc in the reduce call is still active during the f call, which means there are at least two references to acc.

Doesn't reduce force the accumulator to be shared though? Both the reduce and the lambda are holding onto references to acc, which defeats any "single reference" optimizations.

AFAICT progressive AVIF is actually using the "spatial layers" feature of AV1, which was designed to embed multiple resolutions of a video in one bitstream. The limit on four spatial layers comes directly from AV1. The previously decoded layer/progressive-step becomes a reference frame that the next layer can draw from just like eg the temporally-previous frame can be referenced by interprediction.

It has YCgCo too, which should have the efficiency of YCbCr while being lossless.

Being lossless is a function of the whole pipeline from an input image to an output image. The last step being encode to PNG doesn't guarantee anything about losslessness.

AFAIK WebP has never used hardware decoding and no browser uses it for AVIF now.

The DC coefficients in a progressive JPEG are literally an embedded thumbnail (a 1:64 copy of the original) right? Is the question if the second "layer" replaces the thumbnail or layers corrections on top of it? It wasn't clear to me which AVIF does.

Does the difference matter?

I mean Peters himself was German.

I really like oblique projections. Have you seen this neat one from Kerkovits - A Low-Distortion Oblique Map Projection of the World’s Landmasses (2024)?

https://files.catbox.moe/k11gqq.avif

The projection you see with your eyes is actually not that great in a sense. You can only see half of the world at a time and it's quite distorted as you move towards the edge. Also on a computer you should be able to "spin" any map projection, not just a globe.

The article actually maps out the history pretty well. This is basically a recrudescence of the Gall-Peters debate from the 70s, triggered by Boston Public Schools adopting Gall-Peters in 2017. The current wave has taken up the old fight with a less-ugly equiareal map.

It was the second most reported problem on the 2025 Rust survey, after compilation times.

Why is Direct2D so hard to implement in Wine?

General theme: A programming language's native in-memory object format is typically optimized for random access, uniformity, and mutability (fields at fixed offsets, etc). Serialization formats for network or disk tend to be designed explicitly to be more compact. But you can design your own in-memory representation too, with the properties you need.

Oh is that why the page is like 50 megs.

That is the opposite of what GP said. The point is only the total emissions matter, the climate does not care about "fair" allocations over either people or nation-states.

Sounds like it would be easier to hope for a better UPI experience for tourists.

1. The output is a permutation of the input.

2. If the comparison implements a strict total order, the output is sorted according to it.