Scaevolus

4 ms·

1.
by Scaevolus 6d ago
Engineers understand that every technology represents inherent tradeoffs.
2.
by Scaevolus 10d ago
Aside: if you want a source of high quality Piano MIDI recordings, you should check out the MAESTRO dataset: https://magenta.withgoogle.com/datasets/maestro > The dataset contains about 200 hours of paired audio and
3.
by Scaevolus 10d ago
Did you try blender MCP and prompt for low-poly vertex colored assets?
4.
by Scaevolus 12d ago
GPT-6 has a looped transformer / recurrent depth architecture, so it gets some internal CoT reasoning "for free" with no output tokens. This also makes it much harder to monitor its reasoning.
5.
by Scaevolus 23d ago
See Cursor's recent blog post on launching a Github competitor, particularly the architectural history.
6.
by Scaevolus 23d ago
Chrome doesn't use hardware acceleration for decoding any image codecs-- not for JPEG or WebP or AVIF. It has some code for it (using VA-API), but it's disabled by default. Hardware video decoders are generally hard to use for ima
7.
by Scaevolus 1mo ago
"delivering up to 750 output tokens per second and without any quality compromise" seems pretty definitive.
8.
by Scaevolus 2mo ago
- it's easy to prompt an LLM to do it for you, especially if you're targeting a pre-existing runtime like CLR or transpiling to another language. previously you'd have to grind out the parser and compiler for a new language,
9.
by Scaevolus 2mo ago
BitTorrent's bencoding format, used in .torrent files, effectively uses netstrings-- but without the trailing commas, so it uses "5:hello" to represent filenames and similar.
10.
by Scaevolus 2mo ago
It's not really a big problem. You should be storing arbitrary binary strings as blobs (and length() works properly with them), but the underlying encoding is nearly identical: https://www.sqlite.org/fileformat.html#r
11.
by Scaevolus 2mo ago
if you happen to regenerate your protos 5 times in an hour you'll get ratelimited it's a very frustrating system. you should spend a modicum of effort figuring out how to generate your things locally yourself instead, be that dock
12.
by Scaevolus 3mo ago
The absolutely easiest way to write a JIT is to use Javascript and eval() (or "new Function()", which is just eval in a Java-shaped tuxedo). You can quite easily speed up little matching functions, especially arithmetic heavy on
13.
by Scaevolus 3mo ago
Cloudflare's Deepseek V4 Pro prices are 4x more than Deepseek's for input and output tokens, and 100x more for cached input tokens, which is crucial for the tool uses of agents which cause multi-turn conversations.
14.
by Scaevolus 3mo ago
Something like https://www.kythe.io/ ?
15.
by Scaevolus 3mo ago
It's rgb with 3.3 bits per channel, basically 10 bit per pixel color (256 colors is 8bpp).
16.
by Scaevolus 4mo ago
Sqlite can quite easily do 5000+ insert+commits per second on typical NVMe drives. Speed is rarely the constraint that makes it unsuitable for an application.
17.
by Scaevolus 4mo ago
The main thing holding most people back from web-based IDEs is restricted filesystem and tools integrations, but cloud office suites are extremely popular. Google has excellent infrastructure for distributed build and test cycles built into
18.
by Scaevolus 4mo ago
I see you have ALP, but have you tried Chimp128 or Arrow's byte stream split?
19.
by Scaevolus 4mo ago
Apple could easily support eGPUs if they wanted to, but they choose to have vertical integration over fragmentation or usefulness. It's the same as them not supporting OpenGL or Vulkan: they could if they wanted to be a better gaming&#
20.
by Scaevolus 5mo ago
Pangram rates it as 100% AI generated. Heartwarming: even if you die and nobody cares, an AI can write your eulogy!
21.
by Scaevolus 5mo ago
It's not that it's reserving power, but rather that you hit some bottleneck on a 3070 Ti before running into thermal limits-- it's likely limited by either tensor core saturation or RAM throughput. Running the workload with N
22.
by Scaevolus 5mo ago
They are bullshit machines because they do not have an internal mental model of truth like a human does. The flagship models bullshit less, but their fundamental architectures prevent having truth interfere with output. https://p
23.
by Scaevolus 6mo ago
I'm glad there's many teams with automated scans of pypi and npm running. It elevates the challenge of making a backdoor that can survive for any length of time.
24.
by Scaevolus 6mo ago
That's a narrow fair use exception. Many of these open game engines are effectively 1:1 decompilations of the original games, and it would be shocking if they were not effectively copyrighted the same as the original. I don't th
25.
by Scaevolus 6mo ago
Microsoft released a video that covers effectively all of the Xbox One security system, and it's referred to extensively in the talk. The specific methods of glitching don't require any insider knowledge.
26.
by Scaevolus 7mo ago
You're reading more dismissal than I wrote. Most businesses can and many businesses do run efficiently out of shared spreadsheets. Choosing the processes well is the hard part, but there's just not much computational complexity
27.
by Scaevolus 7mo ago
Many engineers get paid a lot of money to write low-complexity code gluing things together and tweaking features according to customer requirements. When the difficulty of a task is neatly encompassed in a 200 word ticket and the implementa
28.
by Scaevolus 7mo ago
JPEG XL also incorporates Chroma from Luma. It's a standard coding tool in modern codecs.
29.
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.
30.
by Scaevolus 8mo 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.

More ›