make3

28 ms·

571.
by make3 3y ago
it's not different. RAG is a way to train embedding stores end to end
572.
by make3 3y ago
This is an exact example of something a vector dbs would be terrible at. Vector dbs work by fetching segments that are similar in topics to the question, so like "Where did <Character> go after <thing>" will retrieve s
573.
by make3 3y ago
comprises means whatever people use it for
574.
by make3 3y ago
one guy deciding to change 90k articles unilateraly is kind of insane indeed.
575.
by make3 3y ago
retro doesn't attend itself, which is a big difference
576.
by make3 3y ago
who's to say.
577.
by make3 3y ago
I'm saying, all of these. Why are they "body moving fast" activated?
578.
by make3 3y ago
It would be nice if we could find what triggers the body's improved condition when doing exercise, and trigger it maximally without having to do hours of repetitive nonsense. It's fine if you like to do it, but a lot of us obviou
579.
by make3 3y ago
adding financial motivation to killing cows will end up in more cows killed. how many cows are used for milk & meat is a money calculation
580.
Conductive Wood for High-Performance Structural EM Interference Shielding(pubs.acs.org)
3 points by make3 3y ago|0 comments
581.
by make3 3y ago
They should make an "upvote" model that tries to predict the amount of likes
582.
by make3 3y ago
why, if you are in America? afaik chatGPT is available everywhere, & there's a huge amount of alternatives
583.
by make3 3y ago
it's interesting that the letter he signed is for large language and language + image (input) models, have nothing to do with this
584.
by make3 3y ago
I feel like the real societal harm of TikTok is the reduction of the attention span
585.
by make3 3y ago
does sandboxing work for things so performance intensive as high end 3d games? (this obviously doesn't apply to low-fi games, including low-fi 3d ones)
586.
by make3 3y ago
yeah LLM for swiftkey is kind of obvious
587.
by make3 3y ago
using plain characters would make the sentences longer & cost much more money to use. that's the idea of byte pair encoding based tokenizers, reduce the average sentence's number of tokens to an optimal (short) size to reduce
588.
by make3 4y ago
probably the best feature
589.
by make3 4y ago
Alpaca uses knowledge distillation (it's trained on outputs from OpenAI models). It's something to keep in mind. You're teaching your model to copy an other model's outputs.
590.
by make3 4y ago
Huggingface Datasets also has an integrated interface to Faiss ! https://huggingface.co/docs/datasets/faiss_es
591.
by make3 4y ago
https://huggingface.co/blog/how-to-generate
592.
by make3 4y ago
this is likely much worse, if copilot uses gpt3.5 turbo or gpt4. who knows though what they're actually using.
593.
by make3 4y ago
it's a game not a shooting simulator? and one of the most successful at that. i'm not sure making it more realistic would achieve anything at all
594.
by make3 4y ago
It just hasn't been prompted or fine-tuned to have the neutral, self effacing personality of ChatGPT. It's doing the pure, "try to guess the most likely next token" task on which they were both trained ( https:/&#x2
595.
by make3 4y ago
Well it's hard to argue against the idea that the final aim of Copilot is to automate as much of software engineering as possible. I don't think that their aim is explicitly to replace humans, but to be as helpful as possible to d
596.
by make3 4y ago
it's totally not the message, weird that you would say this; this is a tool just to make the developers' lives easier
597.
by make3 4y ago
you have no guarantee that that specific information was included in the prompt
598.
by make3 4y ago
"Suddenly linkedin was weirdly silent & devoid of strangely mechanically formulated messages"
599.
by make3 4y ago
it's just a separate vision model. you just have to use a state of the art instance segmentation model, the task shown are really not that hard. it's not "just a library"
600.
by make3 4y ago
they just present this as a cool optimization & aesthetic trick, which it is

More ›