kgeist

46 ms·

751.
by kgeist 3y ago
There's a village here in my area called "Pipes". Every local calls it by this name, but the official name is completely different. There's a legend that there once were 3 old men smoking pipes on their porch who'd
752.
by kgeist 3y ago
Since AI is mostly trained on open data from the Internet, AI is as dangerous as the Internet. Aren't LLMs basically hallucinating search engines which understand natural language? Where's the danger? They can't do anything n
753.
by kgeist 3y ago
>Do you mean using brute force (exact search) and not approximate nearest neighbor search? Yes. >doing an exact search with 100k vectors takes about 1 second. The speed can depend on the language/runtime of your choice and the nu
754.
by kgeist 3y ago
When does a simple linear search become an issue? For example, a cos similarity function which runs even over a million of vectors is surprisingly fast. We can further shard vectors by user (i.e. if they can only search in their own data th
755.
by kgeist 3y ago
We were recently DDOS'd, too. Every time it happens, I wonder, who benefits from it (or even cares about us)? Competitors? Someone's bored? A disgruntled customer? And we'll probably never know.
756.
by kgeist 3y ago
We explicitly generate events (in pseudocode): this.eventDispatcher.dispatch(new SomethingChanged()) It can be anything: just a property change ("AddressChanged"), or something more abstract (i.e. part of business logic),
757.
by kgeist 3y ago
>The solution is to treat these two concepts separately. One subsystem should be used for representing the source of truth, and another should be used for materializing any number of indexed stores off of that source of truth. Once again
758.
by kgeist 3y ago
"Data outlives the application".
759.
by kgeist 3y ago
I wonder what's their plan when a vehicle gets stuck?
760.
by kgeist 3y ago
So it's a character-based model (makes sense, because there's no tokenizer to be found). The model file is 3.3 MB Output after 3 hours of training (loss = 0.4160): hello, the sounds of the rain assistants. In provations whic
761.
by kgeist 3y ago
This project comes from the author of wyhash, which is now the default hashing algorithm in Go, Zig etc. What's lacking is more information about the project. I was able to compile it and just started training on GPU, but I have no ide
762.
by kgeist 3y ago
A cliche is defined as: >a phraseme consisting of components of which none are selected freely and whose usage restrictions are imposed by conventional linguistic usage By their nature, LLMs prefer cliches so it's unsurprising.
763.
by kgeist 3y ago
>We have automated systems in place to detect and resolve this. We’re notified in Discord Isn't Discord hosted on GCP, too? If it goes down, monitoring also goes down?
764.
by kgeist 3y ago
>Two thirds of adults globally (64%) report being happy I think the idea that this reality is a simulation gives people comfort that all their problems are not real and there's a reality somewhere else which is more fair?
765.
by kgeist 3y ago
It's like Occam's razor, we just added another layer of abstraction for no reason.
766.
by kgeist 3y ago
Well, I'm assuming that Youtube has some sort of a script to check if ad traffic is blocked, to detect ad blockers. Sure, you can write some extension which by-passes it, but then they can change their script a week later to do it slig
767.
by kgeist 3y ago
The idea was to VPN only ad traffic (which would be automatically blocked anyway). Proxying video content through Russia would considerably increase latency.
768.
by kgeist 3y ago
Depends on what kind of entertainment. Some forms of it can bring cultures/countries together.
769.
by kgeist 3y ago
Google has been voluntarily blocking ads here in Russia since 2022 due to the sanctions (from what I understand, so that monetization didn't translate into taxes paid to the government). I now live in a completely ads-free world for 2
770.
by kgeist 3y ago
We're moving the goalposts all the time. First we had the Turing test, now AI solving math problems "isn't impressive". Any small mistake is a proof it cannot reason at all. Meanwhile 25% humans think the Sun revolves ar
771.
by kgeist 3y ago
Humans also make mistakes, all the time.
772.
by kgeist 3y ago
For my experiments with new self-hostable models on Linux, I've been using a script to download GGUF-models from TheBloke on HuggingFace (currently, TheBloke's repository has 657 models in the GGUF format) which I feed to a simple
773.
by kgeist 3y ago
GPT3/GPT4 currently moralize about anything slightly controversial. Sure you can construct a long elaborate prompt to "jailbreak" it, but it's so much effort it's easier to just write something by yourself.
774.
by kgeist 3y ago
>According to the paper, Anthropic's superior safety approach was deliberately delaying release in order to avoid “advanc[ing] the rate of AI capabilities progress." Which can end up with China taking the lead. I don't und
775.
by kgeist 3y ago
Just yesterday I saw an example of a person asking GPT what "fluftable" means. The word was invented by their little daughter and they didn't know what it meant. GPT reasoned it was a portmaneau of"fluffy" and &quot
776.
by kgeist 3y ago
I remember playing around in Flash 4 and the site skips versions 2-4. I wonder why they're skipped.
777.
by kgeist 3y ago
I thought lambdas are practical when they're called rarely, with a surge of traffic occasionally? So that you don't pay for the servers when they're unused, and occasionally you can withstand black friday traffic. If a lambda
778.
by kgeist 3y ago
When a system is completely broken for most end users, some companies call it "degraded performance" when it should be, in fact, called "major outage".
779.
by kgeist 3y ago
>But now they have trouble hiring people because there's a tiny pool of people who know that language >In 2023 you've got to have really good reasons for not using C#, or Java, or Python, or TypeScript for building your web
780.
by kgeist 3y ago
Apparently, Brother International is indeed based in Germany

More ›