twotwotwo

9 ms·

31.
by twotwotwo 10mo ago
FWIW, here are (mostly) their agent's tips for other agents from exploring a mostly-new system including tidbits like how to get recent Node: https://s3.us-east-1.amazonaws.com/1FV6XMQKP2T0D9M8FF82-cach... It's ve
32.
by twotwotwo 10mo ago
I have played with it and it's so easy get started with that now I want a quick-project idea as an excuse to use it! I'm sure you've thought of this, but: lots of people have some amount of 'free' (or really: zero
33.
by twotwotwo 10mo ago
So I tried this the other day after Filippo Valsorda, another Go person, posted about it. My reaction was 'whoa, this really makes it easier to start a quick project', and it took a minute to figure out why I felt that way whe
34.
by twotwotwo 10mo ago
Yeah--I wanted a short way to gesture at the subsequent "tasks that are fast for someone but not for you are interesting," and did not mean it as a gotcha on METR, but I should've taken a second longer and pasted what they sa
35.
by twotwotwo 10mo ago
Yeah--it's difficult to go from a benchmark involving the model attempting things alone to the effect assisting people on real tasks because, well, ideally you'd measure that with real people doing real tasks. Last time METR tri
36.
by twotwotwo 10mo ago
I'm conflicted about opining on models: no individual has actually done a large sample of real-world tasks with a lot of models to be able to speak with authority, but I kinda think we should each share our dubiously-informed opinions
37.
by twotwotwo 10mo ago
METR is using hours of equivalent human effort, not actual hours the agent itself spends, so by their methodology, your task might qualify as one where it pulls off much more than 4h of human work. "Human hours equivalent" itself
38.
by twotwotwo 10mo ago
Your version only describes what happens if you do the operations serially, though. For example, a consumer SSD can do a million (or more) operations in a second not 50K, and you can send a lot more than 7 total packets between CA and the N
39.
by twotwotwo 10mo ago
These are potentially complementary approaches. Various innovations have shrunk the KV cache size or (with DSA) how much work you have to do in each attention step. This paper is about hybrid models where some layers' state needs don&#
40.
by twotwotwo 11mo ago
Wanted to note https://issues.chromium.org/issues/40141863 on making the lossless JPEG recompression a Content-Encoding, which provides a way that, say, a CDN could deploy it in a way that's fully transparent to e
41.
by twotwotwo 11mo ago
"What are the consequences of recent, controversial changes in policy?" does not become an irrelevant question simply because you can also think of hypothetical policies.
42.
by twotwotwo 11mo ago
I think there is a lot of progress on efficient useful models recently. I've seen GLM-4.6 getting mention for good coding results from a model that's much smaller than Kimi (~350b params) and seen it speculated that Windsurf based
43.
by twotwotwo 1y ago
There is a largish category of tools now where, unlike in OLTP systems, a big focus is scanning data but quickly (O(n) but with a good constant): Redshift, Trino/Athena, ClickHouse, DuckDB among others. Bloom filter indexing seems like
44.
by twotwotwo 1y ago
Awesome. From curiosity: is Prince's core still written in Mercury? (Looked at old comments.)
45.
by twotwotwo 1y ago
The early users being patent secretaries, then "administrative kind of stuff, typing in trouble tickets," and adoption spreading because people liked it, is kind of cool. That creates different kinds of pressures than a big top-do
46.
by twotwotwo 1y ago
The fast Cerebras thing got me to try the Qwen3 models. I couldn't get them working all that well: they had trouble using the required output format and following instructions. On the other hand, benchmarks say they should be great, an
47.
by twotwotwo 1y ago
Interesting--I'd seen Chinese characters surprise inserted when it was just repeating back input with one provider, but not others. (I'd also occasionally seen tokens surprise-translated to Chinese.) There's a GitHub bug abou
48.
by twotwotwo 1y ago
The tech was surely locked down long ago if they're close to announcing, but just putting the dream out there that AV2 makes next-gen image compression more practical. AVIF's very effective at maintaining OK quality at low bitrate
49.
by twotwotwo 1y ago
Sort of trivial but fun thing: you can fit billions of concepts into this much space, too. Let's say four bits of each component of the vector are important, going by how some providers do fp4 inference and it isn't entirely fal
50.
by twotwotwo 1y ago
They mention it as a block device, and the diagram makes it look like there's one reader. If so, this seems like it has the same function as the page cache in RAM, just saving reads, and looks a lot like https://discord.com&
51.
by twotwotwo 1y ago
Discord: https://discord.com/blog/how-discord-supercharges-network-di... (Somehow the name "SuperDisks" was burned into my brain for this. Although Discord's post does use 'Super-Disks' in a se
52.
by twotwotwo 1y ago
I got a recent ClickHouse and played with it, and looked around GitHub some; replying to this dead thread to update outdated bits of the above. In general: given a huge hash join, you can shrink it (reorder, etc. so fewer rows would need ha
53.
by twotwotwo 1y ago
This is really encouraging! Commented elsewhere in the thread but this was one of the main odd points I ran into when experimenting with ClickHouse, and the changes in the PR and mentioned in the recent video about join improvements ( https
54.
by twotwotwo 1y ago
[Edited to add: looking at comments from saisrirampur and SnooBananas6657, ClickHouse has worked some on joins recently, and there's a very recent a PR around optimizing join order. Cool!] In playing a little with ClickHouse, issues
55.
by twotwotwo 1y ago
> [Hetzner] charges a "one time setup" fee to create that VM. That puts a lot of back pressure on infrastructure decisions and removes any scalability you could otherwise enjoy in the cloud. Hetzner Cloud, then! In the US, $0.5
56.
by twotwotwo 1y ago
It is kind of wild that most coding tasks are editing tasks, and we humans care a lot about code editing tools, but automated tools use code generation for editing where a valid block must be generated top-to-bottom in one go. Fixing a
57.
by twotwotwo 1y ago
trane_project is selling a $20/mo subscription or $1000 perpetual license to their own reading program and folks should read this and their other comments aware of that context. It's disappointing to tell a personal story, come ba
58.
by twotwotwo 1y ago
For what it's worth, a pivotal moment for keeping reading going after the lessons was when my partner picked up a comic book at a library event. For a few weeks after the end of the lessons, reading time had been traditional early re
59.
by twotwotwo 1y ago
For any parents of small kids here, I have to mention the book Teach Your Child to Read in 100 Easy Lessons. We went through it while my kid was in kindergarten, and after that, I absolutely believe what I've heard from parents who did
60.
by twotwotwo 1y ago
It was insightful of him and there absolutely are similar strains of hype today! He also both points out the silliness of the pitch for the hype-y tech of the day and calls that same tech a tool some might use with good reason. That doe

More ›