Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
ww520
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
26 ms
·
361.
▲
by
ww520
2y ago
Wasn't it how people steal gasoline back in the days of gas caps not being locked? Guess it's a lost art.
362.
▲
by
ww520
2y ago
This speaks the importance of locally hosted models.
363.
▲
by
ww520
2y ago
Monad provides a way to wrap types of data to process and chain them uniformly.
364.
▲
by
ww520
2y ago
Yes. Limewire came out a bit later. All the apps with the same protocol could talk to each other. Initially mine was open sourced as well. I close sourced it after selling it to a company.
365.
▲
by
ww520
2y ago
In cities with less busy police departments, police actually do investigative work. I had police came by couple times asking for security footage for investigation nearby. One time it’s for a family member of a neighbor down the street “sto
366.
▲
by
ww520
2y ago
Ah good time. Bring back memory. My involvement with that period was I wrote a p2p file sharing app using the Gnutella protocol, an alternative to the Napster. It was a fun time.
367.
▲
by
ww520
2y ago
The word embeddings computed from the hex values and the car dealership's inventory ID's probably have close similarity in Google's vector db.
368.
▲
by
ww520
2y ago
Good explanation. Always good to read up on some bit tweaking once a while.
369.
▲
by
ww520
2y ago
It can run on Wine on Linux. Best of both worlds!
370.
▲
by
ww520
2y ago
Foobar2000 has a very good tag editor. I especially like how it can batch-sync the title tags from the filenames (via Properties -> Tools -> Automatically Fill Values). And the other way around to rename files using their tags (via
371.
▲
by
ww520
2y ago
RAG is augmenting the llm generation with external data. How the external data is retrieved is irrelevant. A search is not necessary. Of course you can do a search on the related tables with regard to the question to narrow down the table l
372.
▲
by
ww520
2y ago
RAG is not just traditional search. It's any augmented data that can be fed to the LLM. The most useful and verifiable RAG setup I've seen is hooking up a RDBMS and LLM, and asking querying questions in English to retrieve the ta
373.
▲
by
ww520
2y ago
Dug up more generated drawings. https://i.imgur.com/lw7HP5k.jpeg (Forest) https://i.imgur.com/2okI9da.jpeg (Conversation) https://i.imgur.com/w1cQDcO.jpeg (Lava flow)
374.
▲
by
ww520
2y ago
I've dabbled in generative art a while back. Never got the chance to formalize it and make it repeatable. Here're some samples. https://i.imgur.com/U9MPrIc.jpeg (Stormy Sea) https://i.imgur.com/o
375.
▲
by
ww520
2y ago
I imagine having an agent set up with specific RAG context to solve a specific problem and having another with a different RAG context to solve a different problem can be useful.
376.
▲
by
ww520
2y ago
Actually it’s clever not dumb. It partial evaluates to the first change then stops. It does the job and no more. Very insightful. I’m stealing this idea. :)
377.
▲
by
ww520
2y ago
Yes, when I saw the immutable path is cloned and appended a key, I knew the benchmark was a strawman.
378.
▲
by
ww520
2y ago
I've seen a lexer/parser scheme that encodes the lexer token type along with the token file location information into a u64 integer, something like struct Token { token_type: u8, type_info: u8, token_
379.
▲
by
ww520
2y ago
Nice article. Good explanation and good guidelines. Sometimes the evenly split windows don't cut it. I put the following function in the init.el file and bind "C-x 7" to it to adjust the current window and its peer to 70% a
380.
▲
by
ww520
2y ago
Same here. My first job out of college was at a database company. Queries at the client side of the client-server based database were slow. It was thought the database server was slow as hardware back then was pretty pathetic. I traced
381.
▲
by
ww520
2y ago
The 4% or 3% as OP put it withdrawal rate accounts for inflation.
382.
▲
by
ww520
2y ago
For compile time control and meta programming, I found Zig's comptime system to be the best. It blends with the regular syntax seamlessly and solves several problems at the same time. It's one of the best things came out of lang
383.
▲
by
ww520
2y ago
This is an amazing game. Nice work! May be adding mouse control? Turning with arrow keys are kind of clunky.
384.
▲
by
ww520
3y ago
MMA has a way to humble anyone. It's a great way to train for body and mind.
385.
▲
by
ww520
3y ago
The earthquake might cause unexpected damage. The Crystal Springs reservoir sits right on the San Andreas fault line. The reservoir supplies drinking water to the SF county and San Mateo county, all the way to Palo Alto. An earthquake might
386.
▲
by
ww520
3y ago
Ok. So it's indeed restricted to 8 core (1 thread per core). Then the benchmark above using 16 threads was not really a fair comparison.
387.
▲
by
ww520
3y ago
Good to know. I didn’t know the contest has no limit on hyperthread.
388.
▲
by
ww520
3y ago
Also this uses 16 threads while the contest restricts to running in 8 cores. Needs to compare the benchmarks in the same environment to make a fair comparison.
389.
▲
by
ww520
3y ago
The key can be up to 100 bytes. The key in the hashmap is in cache. The key being compared is from main memory. Basically the whole file in memory is being gone through and compared. With 1 billion rows it’s about 16GB data in memory for a
390.
▲
by
ww520
3y ago
Actually on modern systems it’s pure compute plus memory bounded. For this problem hashmap lookup is the biggest bottleneck. Parsing takes time away as well but not as bad as hashmap.
More ›