derefr

101 ms·

781.
by derefr 2y ago
This is an important point: what any professional investor (i.e. someone with a fiduciary duty to their company to attempt to make money through investing, rather than just gambling) says in public about their investments, is going to be fa
782.
by derefr 2y ago
...but I wasn't describing a "human-assisted interaction interface and workflow." I was describing a different way for an AI to do things "inside its head" in a feed-forward span-of-a-few-seconds inference pass.
783.
by derefr 2y ago
I would argue the opposite — image generation is the clear loser. If you've ever tried to do it yourself, grabbing a bunch of LoRAs from Civitai to try to convince a model to draw something it doesn't initially know how to draw —
784.
by derefr 2y ago
What's your example of "real work"? Most "well-known-name" open-source ML models, are very much "base models" — they are meant to be flexible and generic, so that they can be fine-tuned with additional t
785.
by derefr 2y ago
> I’ve have literally never once in my life ever written a comment as long as yours I get this sort of reply almost every time I write a "divisive" comment. I write a lot, I admit; but it doesn't take me particularly long
786.
by derefr 2y ago
> trade shows, conferences, introduction meetings, business mixers or meet snd greets Alright, sure. I will admit that this is the "obvious"-at-first-glance use for business cards. I thought this way myself, until the rubber hi
787.
by derefr 2y ago
> I just want to know what number to call when I want that service. I have literally never once in my life been in the situation of wanting to hire someone to provide me a service after the first time I met that person, in person. Heck,
788.
by derefr 2y ago
But, whether in a personal or a business context, I don't do... either of these things? Mostly, if I'm meeting someone for the first time in person , then it's either through friends, or at an event. If it's through fri
789.
by derefr 2y ago
> The main signals on Search have been AI/ML for >5 years now. Say more? "Passing the entire web through ML-model inference to generate 'signals' — IQL expressions? — at the same global-concurrent throughput as pre
790.
by derefr 2y ago
How else would you expect a green-threaded runtime with multiple scheduler threads to spawn a managed subprocess (think: Erlang port-program) that has access to the runtime's file descriptors, unix domain sockets, etc?
791.
by derefr 2y ago
As a sci-fi author, I love thinking about this kind of "what-if biology." I presume, if such a phenomenon was possible, that it would work the same way electric eels work: by electrochemical (dis)charge. You don't need the wh
792.
by derefr 2y ago
> as opposed to all the shades of red and blue that are missing because really saturated red and blue primaries are too dim (per unit of energy) to use. Would another way to put that be, that the chromaticity diagram could keep going s
793.
by derefr 2y ago
"Becoming"? Dumb overzealous bots (e.g. broken API clients that retry immediately forever; scrapers that want your 5PB dataset right now ; etc) that won't go away — so you have to confuse/stall/crash them to keep y
794.
by derefr 2y ago
I presume that in terms of the OpenAI Assistants API (i.e. what you’d be using if you have a specific “use case”), search will be just another built-in tool (like the code interpreter) that an Assistant can be told to invoke. (And “SearchGP
795.
by derefr 2y ago
These days, you can't even pick out a brand-name product from a brick and mortar store and be sure of getting the real thing. People will buy something from a store, buy a fake version of that thing on AliExpress, swap the fake into
796.
by derefr 2y ago
Nothing on Amazon is going to work like books on Amazon do. Amazon is, originally and primarily, a bookstore. Amazon has long-established direct contracts with all book publishers (and in fact, a stranglehold over those publishers.) Any tim
797.
by derefr 2y ago
Due to the way tokenization usually works with LLMs (using BPE — Byte Pair Encoding), there's actually usually already a 256-element embedding within the token-space that represents "raw bytes." You could say that this 256-
798.
by derefr 2y ago
> Unicode is definitely a prescriptive standard. Code point 65 is a capital A because Unicode says so -- any such standard is necessarily prescriptive. You're missing the point. Unicode is descriptive like a dictionary is descriptiv
799.
by derefr 2y ago
Unicode is descriptive, not prescriptive. The point of any glyph at all being allocated a Unicode codepoint, is that the glyph already exists in documents somewhere (either real paper documents, or digital documents in some other legacy enc
800.
by derefr 2y ago
Don’t forget machine-translated texts, where until ~2017 the translation was likely done by something much dumber / semantically lossy than an LLM, and after 2017 was basically done by an early form of LLM (the Transformers architectur
801.
by derefr 2y ago
Given current SOTA, no, they don’t. But there’s no reason why they couldn’t — just capture the vectors of some of the earlier hidden layers during the RAG encoder’s inference run, and append these intermediate vectors to the final embeddi
802.
by derefr 2y ago
> adding a contextual title, summary, keywords, and questions That's interesting; do you then transform the question-as-prompt before embedding it at runtime, so that it "asks for" that metadata to be in the response? Beca
803.
by derefr 2y ago
Compare/contrast: DOOM-the-engine is open source ( https://github.com/id-Software/DOOM ), even though DOOM-the-asset-and-scenario-data is not. While you need a copy of DOOM-the-asset-and-scenario-data to "use D
804.
by derefr 2y ago
Sounds very military… and yet I see no mention of a suite of remotely-armable tamper sensors built into the hardware, that would enable automatic scorched-earth wiping of sensitive data if their guys manage to shoot your guys and march
805.
by derefr 2y ago
No, it's the difference between owned housing vs renting from a landlord who is also your boss in a company town , where the landlord has a vested interest in e.g. preventing you from using your apartment to also do work for a competi
806.
by derefr 2y ago
People do do it, though. Especially poor people, who might not use their personal computers very often. Also, many people don't own a separate "personal" computer in the first place. Especially, again, poor people. (I know
807.
by derefr 2y ago
> Hmm I think this is unneeded, vmlinux already has all the code. Yeah, I was just thinking about it as a way to reduce the scope of the "preload" step of symbolic interpretation, for the case where you want to work with semi-s
808.
by derefr 2y ago
Tangent — a question you didn't ask, but I'll pretend you did: > If employers allowed employees to "bring their own devices", and then didn't force said employees to run MDM software on those devices, then how
809.
by derefr 2y ago
> What does “our computer” mean when it is not owned by you, but issued to you to perform a task with by your employer? Well, presuming that: 1. the employee is issued a computer, that they have possession of even if not ownership (i
810.
by derefr 2y ago
The thing is that even today, where it's "easy", still nearly nobody is writing code to directly support the six different credit card types. Instead, businesses mostly rely on payment processors. In a world where it's

More ›