|
PIZZA.BAT posted:i'm hearing vector databases are the new hotness that's getting all the vc attention finally, scalable databases!
|
# ? Mar 20, 2024 16:46 |
|
|
# ? May 2, 2024 17:52 |
|
Subjunctive posted:are vector databases now doing LLM-style random walk everywhere? I thought it was just nearest-neighbour with lots of dimensions no real randomness (not that central to llm's either after all, though it is involved), but if you're starting from text and want some high-dimensional point in 2024 i think you probably *want* something very llm'ish.
|
# ? Mar 20, 2024 17:34 |
|
unlike chatbots, vector databases are actually useful. you can basically use them to do semantic search. you query for butts, it returns both farts and poops, because those vectors have high cosine similarity to your query. the embedding into vectors is where the machine learning magic happens. but it doesn't do generative chatbot bullshit, it just returns the search results. people are then using these results to submit the original query to chatgpt with additional context in the prompt (they call this retrieval-augmented generation, as though it was even deserving of a name), but you don't have to do that part
|
# ? Mar 20, 2024 18:34 |
|
DELETE CASCADE posted:people are then using these results to submit the original query to chatgpt with additional context in the prompt (they call this retrieval-augmented generation, as though it was even deserving of a name), but you don't have to do that part so now they're sending prompts through a thesaurus. wonderful
|
# ? Mar 20, 2024 19:21 |
|
redleader posted:so now they're sending prompts through a thesaurus. wonderful bdid did it first
|
# ? Mar 20, 2024 20:24 |
|
christ, i'm still actively hating mongodb and now they've got them newfangled vector database things i need to be furious about too? also i like postgres more than mysql in large part because it actually treats your data as if it's valuable. mysql lets you (or used to let you, maybe it changed) do poo poo like dropping tables that still had foreign key references on them if you just added something to the ddl statement that was basically the equivalent of "trust me bro", whereas postgres will firmly tell you to get hosed if you try bullshit like that i will also never not be angry at mysql for using the "long" type in their c client library api which means it's 32 bit on x86 and x64 windows, while it's 32 bit on x86 linux but 64 bit on x64 linux. combine that with having to pass long* values to get query results and you're in for some exciting times if you're writing code that should run on all four of those platforms and someone just carelessly shoves a cast from int to long in the wrong spot; doubly so if they're developing on a windows machine since it'll work just fine there
|
# ? Mar 21, 2024 13:46 |
|
Deep Dish Fuckfest posted:i will also never not be angry at mysql for using the "long" type in their c client library api which means it's 32 bit on x86 and x64 windows, while it's 32 bit on x86 linux but 64 bit on x64 linux lmao, incredible
|
# ? Mar 21, 2024 22:07 |
|
Deep Dish Fuckfest posted:also i like postgres more than mysql in large part because it actually treats your data as if it's valuable. mysql lets you (or used to let you, maybe it changed) do poo poo like dropping tables that still had foreign key references on them if you just added something to the ddl statement that was basically the equivalent of "trust me bro", whereas postgres will firmly tell you to get hosed if you try bullshit like that postgres won't save you from yourself if you use the appropriate incantation (username post combo!!!) code:
|
# ? Mar 21, 2024 22:14 |
|
rotor posted:finally, scalable databases! it's a vector database, not a scalar one
|
# ? Mar 22, 2024 03:12 |
|
DELETE CASCADE posted:unlike chatbots, vector databases are actually useful. you can basically use them to do semantic search. you query for butts, it returns both farts and poops, because those vectors have high cosine similarity to your query. the embedding into vectors is where the machine learning magic happens. but it doesn't do generative chatbot bullshit, it just returns the search results. people are then using these results to submit the original query to chatgpt with additional context in the prompt (they call this retrieval-augmented generation, as though it was even deserving of a name), but you don't have to do that part Oh hey this is what I’m doing at work. We’re using a vector DB to store a bunch of insights extracted from driving data for the purposes of finding it later to feed into simulators for self driving cars. So if your self driving algorithm is messing up when it’s wet out and you’re approaching an overpass you could type in “overpass while raining” or something like that and get some really accurate results. It’s pretty incredible how well it works We demoed it at IAA last year and the CEO stopped by; the scientists did all the cool vector extraction stuff but were gonna demo either Jupiter notebooks or some hacked up voxel install when I came in to make a little web app, hosted it on ecs, set up a login page blah blah blah… it took a solid month but it was really fun. It was so well received it changed the direction of our project… which I think upset some PMs because when they got back i was not mentioned in the org wide project acknowledgements lol. I’m not mad though please don’t tell the newspaper I got mad Edit to anonymize it a bit ADINSX fucked around with this message at 04:15 on Mar 22, 2024 |
# ? Mar 22, 2024 03:36 |
|
DELETE CASCADE posted:postgres won't save you from yourself if you use the appropriate incantation (username post combo!!!) well yeah no poo poo you're literally asking postgres to delete dependencies too, which is fine, and it means that whatever invariants you're enforcing in what's left in your db will still be correct. mysql will let you drop a table without cascading the drop to dependencies! as in you have constraints that refer to a table that doesn't exist anymore. because who gives a poo poo about data integrity. also i just looked at the mysql doc to see if things changed and lol quote:The RESTRICT and CASCADE keywords do nothing. They are permitted to make porting easier from other database systems. and agreed on the transactional ddl; makes upgrading stuff way less stressful
|
# ? Mar 22, 2024 03:43 |
|
Hed posted:it's a vector database, not a scalar one vectors are easily scalable and dont leave jaggies
|
# ? Mar 22, 2024 04:46 |
|
Deep Dish Fuckfest posted:mysql will let you drop a table without cascading the drop to dependencies! as in you have constraints that refer to a table that doesn't exist anymore. because who gives a poo poo about data integrity. i was not aware of this lol
|
# ? Mar 22, 2024 05:42 |
|
do they not maintain the schema in their own internal system table? wtf? how can your ddl not be transactional?
|
# ? Mar 22, 2024 12:49 |
|
bold of you to assume there's some internal system table
|
# ? Mar 22, 2024 14:02 |
|
just follow our Devs and set no keys or relationships in the first place
|
# ? Mar 22, 2024 14:31 |
|
when i first started learning php and mysql in 2003, I had no idea that you could even define foreign keys as part of the db schema itself. everything was built on raw queries and hope. so many orphans in those days.
|
# ? Mar 22, 2024 14:39 |
|
DELETE CASCADE posted:i was not aware of this lol username post combo
|
# ? Mar 22, 2024 14:47 |
|
Deep Dish Fuckfest posted:mysql will let you drop a table without cascading the drop to dependencies! as in you have constraints that refer to a table that doesn't exist anymore. Is that also true for deleting individual rows? As in, in table "butt" I have a row with buttId1 and in table "fart" I have a row with fartId 1 and a foreign key buttId 1, and I can delete the row in butt no-problemo, leaving the fart row with a dangling reference?
|
# ? Mar 24, 2024 14:05 |
|
PIZZA.BAT posted:do they not maintain the schema in their own internal system table? wtf? how can your ddl not be transactional? there is a difference between “the schema text is stored transactionally”, which is indeed simple, and “the effects of changing the schema are handled transactionally”, which is a fair bit more involved
|
# ? Mar 24, 2024 14:13 |
|
I haven't read the thread at all but i'm using sqlite3 atm and it's p deece. It's kind of weird that the schema is stored as SQL statements so half of the ALTER TABLE stuff won't work but hey ho
|
# ? Mar 24, 2024 16:01 |
|
Ocean of Milk posted:Is that also true for deleting individual rows? As in, in table "butt" I have a row with buttId1 and in table "fart" I have a row with fartId 1 and a foreign key buttId 1, and I can delete the row in butt no-problemo, leaving the fart row with a dangling reference? what happens depends on the foreign key definition and its "on delete" clause, but i don't think even mysql lets you have dangling references like that. the actual details of the behavior probably depends on the storage engine you're using, like many things in mysql. that said i could've sworn there was a way to get the behavior you're describing, but it's been a long time since i used mysql and i can't remember. the closest would probably be adding the "ignore" modifier to the delete statement where rows that would normally fail a constraint and cause execution to abort will just get skipped instead, ensuring you have no idea what exactly it is that you just deleted why the gently caress you'd want that behavior let alone actually allow it is a mystery to me
|
# ? Mar 24, 2024 20:14 |
|
because bossman is breathing down my neck and IDGAF anymore ok fine here its gone are you fukin happy now its gone now goddamn <bossman leaves> gently caress. what did i just break. ahhhh fuckit will worry out it when something happens.
|
# ? Mar 25, 2024 13:19 |
|
looks like mariadb has System-Versioned Tables now which is nice https://mariadb.com/kb/en/system-versioned-tables/
|
# ? Mar 27, 2024 05:40 |
|
my favorite database is a text file I store next to the executable
|
# ? Mar 27, 2024 18:25 |
|
now i'm wondering if sqlite has an option to store data in human readable format
|
# ? Mar 27, 2024 18:37 |
|
Deep Dish Fuckfest posted:now i'm wondering if sqlite has an option to store data in human readable format Yeah you just run strings on the db file
|
# ? Mar 27, 2024 18:40 |
|
i was just on a call with one of my customers where our machine learning guy was walking them through various use cases to help them brainstorm things that could be applicable for them. the discussion meanders around for a bit and we hit the inevitable, 'how many records does your dataset contain' question. they answered fifty. we asked, 'fifty... million? billion? what?' no. just fifty. they bought an enterprise license from us to store fifty documents props to my machine learning guy for gently taking that and going, 'ok so you may run into an issue where the model overcorrects here and blah blah blah' instead of laughing fifty documents
|
# ? Mar 27, 2024 19:18 |
|
maybe they're big documents?
|
# ? Mar 27, 2024 19:38 |
|
PIZZA.BAT posted:i was just on a call with one of my customers where our machine learning guy was walking them through various use cases to help them brainstorm things that could be applicable for them. the discussion meanders around for a bit and we hit the inevitable, 'how many records does your dataset contain' question. they answered fifty. we asked, 'fifty... million? billion? what?' im a busy, handsome man with no time to figure out where to put these fifty documents. Heres a check, figure it out nerdlingers.
|
# ? Mar 27, 2024 19:52 |
|
each document is one entire SQLite database containing one billion rows
|
# ? Mar 27, 2024 20:24 |
|
and one column
|
# ? Mar 27, 2024 20:25 |
|
rotor posted:im a busy, handsome man with no time to figure out where to put these fifty documents. Heres a check, figure it out nerdlingers. wisdom
|
# ? Mar 27, 2024 20:48 |
|
Deep Dish Fuckfest posted:now i'm wondering if sqlite has an option to store data in human readable format the SQLite disk format is readable by humans if that human is motivated enough
|
# ? Mar 27, 2024 20:59 |
|
that's true of every file format
|
# ? Mar 27, 2024 21:10 |
|
now you’re seeing the truth
|
# ? Mar 27, 2024 21:36 |
|
how to read sqlite as a human: step 1. install sqlite
|
# ? Mar 27, 2024 23:34 |
|
Subjunctive posted:now you’re seeing the truth reminds me of that time i got curious how "real" file formats actually look like and read the specs for stuff like elf and dwarf and pdf and who knows what else my conclusion was that everything is garbage and i'll just keep dumping raw memory to disk like god intended without giving a gently caress about things like padding or versioning, and most certainly not about endianness because those big endian freaks shouldn't be allowed to touch my data in the first place
|
# ? Mar 27, 2024 23:52 |
|
pdf is a trip for sure
|
# ? Mar 27, 2024 23:54 |
|
|
# ? May 2, 2024 17:52 |
|
fun fact: the first programming language i learned as an adult was postscript
|
# ? Mar 28, 2024 00:18 |