Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
antirez
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
21 ms
·
301.
▲
by
antirez
11mo ago
Near perfect recall VS fp32, not in absolute terms: TLDR, it's not int8 to ruin it, at least if the int8 quants are computed per-vector and not with global centroids. And also, recall is a very illusionary metric, but this is an argume
302.
▲
by
antirez
11mo ago
Fascinating things about other small worlds applications, very far from what I do, that I did ignore. Thanks.
303.
▲
by
antirez
11mo ago
Hi! I worked with product quantization in the past in the context of a library I released to read LLMs stored in llama.cpp format (GUFF). However, in the context of in-memory HNSWs, I found them to make a small difference. The recall is alr
304.
▲
by
antirez
11mo ago
Well TLDR I'm an idiot :D I write blog posts into TextEdit, just the white page to fill with text. Normally this is fine as I end writing and publish the blog post. This time after writing it I left it there for weeks: I wanted to refi
305.
▲
by
antirez
11mo ago
Thanks! Appreciate your words.
306.
▲
by
antirez
11mo ago
There is an entire section of the post about that. I believe that's more the illusion of a problem because of product design issues than a real challenge since far results that match the filter are totally useless.
307.
▲
by
antirez
11mo ago
I know that the OP proposal sounded absurd :D But, can you believe that there are many folks that learned Italian following my channel? I'm shocked but this happened multiple times in the latest months. The fact is that Italian is incr
308.
▲
by
antirez
11mo ago
Hi! I'm perfectly capable of doing the same series in English, not that I'm so skilled with other languages, but enough to do it: you can find many programming videos in English on my channel. However: it will never be at the sa
309.
▲
by
antirez
11mo ago
In my YouTube C language course in the latest episodes I'm implementing a Toy FORTH interpreter: I believe FORTH is nice even just for the educational value it contains, and also, together with Lisp, it shows another way to perform c
310.
▲
by
antirez
11mo ago
So many hours of fun watching videos over the years, without the constant fear that some odd codec would make the experience impossible. Thanks, VLC!
311.
▲
by
antirez
11mo ago
It works with image embeddings too: https://youtu.be/r6TJfGUhv6s?si=_LC0d4Mwyw18c53B
312.
▲
by
antirez
11mo ago
"But almost all programs have paths that crash, and perhaps the density of crashes will be tolerable." This is a very odd statement. Mature C programs written by professional coders (Redis is a good example) basically never cras
313.
▲
by
antirez
11mo ago
Sane async does not need io_uring. All the non-linear-time select(2) alternatives are good enough.
314.
▲
by
antirez
11mo ago
Redis Vector Sets, my work for the last year, I believe address many of such points: 1. Updates: I wrote my own implementation of the HNSW with many changes compared to the paper. The result is that the data structure can be updated while i
315.
▲
by
antirez
11mo ago
It's not a module, it is part of every new Redis version now. Well, actually: it is written in the form of a module and with the modules API in order to improve modularity of the Redis internals, but it is a "merged module"
316.
▲
by
antirez
1y ago
Basically often it could be better to have an incident in a nuclear plant than in some building construction site... But the attention delta is incredibly high. 300 CPM sounds low, I hope they will be fine.
317.
▲
by
antirez
1y ago
This should be "pixels are (maybe) a better representation than the current representation of tokens". Which is very different. Text is surely more information dense than the image containing the same text, so the problem is findi
318.
▲
by
antirez
1y ago
"SVG generation as a quality proxy" No need to read further.
319.
▲
by
antirez
1y ago
To check for consistency in the reasoning steps in the presence of a correct reply, to evaluate the actual LLM performances, is a fundamentally misleading idea. Thinking models learn to do two things: 1. to perform sampling near the problem
320.
▲
by
antirez
1y ago
This object makes sense to me only if, even if there is a display, which is fundamentally different than tracing line with the CRT raster, at least that original process is simulated. If the lines buffer is rendered just with a line drawi
321.
▲
by
antirez
1y ago
Yep the STM32 has certain advantages, even if they are usually complicated to program. With the Flipper Zero I programmed a low level timer interrupt and had to setup don't know how many registers... And the documentation is lacking. A
322.
▲
by
antirez
1y ago
I'm not sure how it is possible to compete against the RP2040/RP2050 at this point: quality, great programming environment, awesome hardware specs, cheap.
323.
▲
by
antirez
1y ago
Lua that isn't too old is not compatible with old Lua, unfortunately.
324.
▲
by
antirez
1y ago
Yep, however people don't configure things properly so many years ago I introduced a middle ground between not listening to * (which makes things harder for users in actual deployment systems) and leaving the server exposed, that is: p
325.
▲
by
antirez
1y ago
That's exactly how it works in most fields that are not purely engineering but where the space of design solutions to do X is huge. Architecture, software development, ...
326.
▲
by
antirez
1y ago
What bambulab printers that have the "A1" series nozzle, replacing one is a matter of 3 seconds, just by naked hands. 10x times faster than in the Prusa world (Nextruder included). Being this the single most important thing a prin
327.
▲
by
antirez
1y ago
EDIT: Sorry didn't realize it was for the specified range! So I doubt my comment makes sense in light of that.
328.
▲
by
antirez
1y ago
Notably, still less than in any country in the European Union: given the lifestyle, is this a matter of the health care system, I guess? Anyway given that random EU folks live longer without switching to 1800 lifestyle, looks like there are
329.
▲
by
antirez
1y ago
Great article not mentioning local backups were already available and what this is about. The state of affairs in iOS vs Android of the past feature and the next one. Details of all the kind are missing. WTF.
330.
▲
by
antirez
1y ago
The difference is enormous: in one case, you don't need any centralized N entities, just a big percentage of the network, whoever wants to participate, runs the protocol and there are no 50 institutions / companies that can block
More ›