Scaevolus

6 ms·

31.
by Scaevolus 8mo ago
Hardware video decoding APIs often have significantly more latency than software decoders, to the point that it's a noticeable several hundred milliseconds of delay. If they have this delay, they're unusable for images.
32.
by Scaevolus 9mo ago
You have that backwards. GFS was replaced by Colossus ca. 2010, and largely functions as blob storage with append-only semantics for modification. BigTable is a KV store, and the row size limits (256MB) make it unsuitable for blob storage.
33.
by Scaevolus 9mo ago
Most companies doing CGI work, both in games and movies, are quite open about their technical details. The whole industry relies on pooled research and development. While the actual code is typically confidential, publishing information ser
34.
by Scaevolus 9mo ago
When LLMs predict the next token, they actually produce a distribution of the probability of each of the possible next tokens, and the sampler chooses one of them, and not necessarily the most likely one! If instead you run LLM prediction a
35.
by Scaevolus 9mo ago
Client-side frame extraction is far too slow to be usable for large volumes of data. You want to precompute the contact sheets and serve them to users. You can encode them with VP9, mux to IVF format, and use the WebCodec API to decode them
36.
by Scaevolus 9mo ago
GGML is another neat ML abstraction layer, but I don't think much work has been dedicated to the Windows port.
37.
by Scaevolus 9mo ago
Shipping GPU-accelerated ML models in games looks difficult, are there any major examples other than vendor-locked upscaling like DLSS or FSR? (Yep! https://cod.ifies.com/voxel-visibility/ )
38.
by Scaevolus 9mo ago
Are the LLMs run on-device, or does this use cloud compute? (Off-topic AMA question: Did you see my voxel grid visibility post?)
39.
by Scaevolus 9mo ago
The pre-machine-learning formulations of AI focused on symbolic reasoning through the dual problems of search and logic. Many problems can be reduced to enumerating legal steps, and SAT/SMT/ASP and related systems can churn throug
40.
by Scaevolus 9mo ago
Yes, this is the major challenge with solving them with SAT. You can make your solver check and reject these horseless pockets (incrementally rejecting solutions with new clauses), which might be the easiest method, since you might need ite
41.
by Scaevolus 9mo ago
The site uses Answer Set Programming with the Clingo engine to compute the optimal solutions for smaller grids. Maximizing grids like this is probably NP-hard. Note that traditional SAT and SMT solvers are quite inefficient at computing f
42.
by Scaevolus 9mo ago
Apple had soldered DDR ram for a long time that was no faster than any other laptop. It's only with the Apple Silicon M1 that it started being notably higher bandwidth.
43.
by Scaevolus 9mo ago
2026 YOLOTD!
44.
by Scaevolus 9mo ago
https://x.com/vxunderground/status/2005008887234048091 Here's the word on the internet streets: - THE FIRST GROUP of individuals exploited a Rainbow 6 Siege service allowing them ban players, modify inventory
45.
by Scaevolus 9mo ago
Self-injecting feels like a scary, painful, dangerous procedure and becomes completely boring by the third repetition.
46.
by Scaevolus 9mo ago
Auto-vectorization is consistently one of the least predictable optimization passes, which is rather awful, since when it doesn't trigger your functions are suddenly >3x slower. This drives people to more explicit SIMD coding, fro
47.
by Scaevolus 10mo ago
Moonlight is mostly designed to stream your gaming desktop to a portable device or your TV at minimal latency and maximum quality within a LAN. For that, 40Mbps is quite reasonable. It's obviously absurd for mundane VNC/productivi
48.
by Scaevolus 10mo ago
Note that WWV is a good 60 miles NNE of Boulder on the outskirts of Fort Collins.
49.
by Scaevolus 10mo ago
You can use fiber optics as an optical delay line too! About 60KB/km at 100Gbps. https://en.wikipedia.org/wiki/Delay-line_memory
50.
by Scaevolus 10mo ago
Especially given how the response time doesn't matter much here! If you're just looking at 2-character license plates, that's 676/5=136 requests to check them all, and you could easily space that out to something like
51.
by Scaevolus 10mo ago
Is the large significands table (~19KB) an issue for cache contention in practical use, or are only a small range of significands generally accessed?
52.
by Scaevolus 10mo ago
SDL3 is kind of the intersection of features found in DX12/Vulkan 1.0/Metal: if it's not easily supported in all of them, it's not in SDL3-- hence the lack of bindless support. That means you can run on nearly every devi
53.
by Scaevolus 10mo ago
This could probably be implemented as an expensive comment-driven lint during compilation.
54.
by Scaevolus 10mo ago
He went even farther, decapping the CPU and reverse engineering its entire workings to understand how each "undocumented" opcode functioned: https://www.pagetable.com/?p=517
55.
by Scaevolus 10mo ago
Note that ffprobe can output JSON which is much easier to handle than CSV. I have this snippet in my bashrc: ffpj() { for f in "$@"; do ffprobe -v quiet -print_format json -show_format -show_streams "$f";
56.
by Scaevolus 10mo ago
They saw errors and decided to do a second rollout to disable the component generating errors, causing a major outage.
57.
by Scaevolus 10mo ago
> First, individuals who choose vaccination may differ from those who do not, potentially introducing confounding bias. It's very hard to interpret this data given the massive confounder of "antivaxxers are suspicious of health
58.
by Scaevolus 10mo ago
> Disabling this was done using our global configuration system. This system does not use gradual rollouts but rather propagates changes within seconds to the entire network and is under review following the outage we recently experience
59.
by Scaevolus 10mo ago
Oops, I guess that one is small enough to use Zoekt. Note how different the interface is than Android and Chromium's code search.
60.
by Scaevolus 10mo ago
None of Google's public source search engines (android, chromium, bazel) use Zoekt. They use Google's web indexing technology adapted for trigrams, which was mostly developed to support their massive internal monorepo, but exposed

More ›