What are the advantages and potential challenges of combining dense vectors, sparse vectors, and full-text search in a hybrid retrieval method, as implemented in Infinity v0.2, and how does this approach compare to traditional vector search or other retrieval methods?
A noticeble work to demostrate the effectiveness of hybrid search is blended rag by IBM research (https://arxiv.org/abs/2404.07220), which has shown that 3-way hybrid search could achieve STOA over multiple evaluation datasets.
And also, we've reproduced the results of blended rag, as shown in this article. Additionally, blended rag + colbert based reranker could have a much better results.
The major challenges are how to implement and manage such many indices within single database. That's why we build this database start from scratch. Infinity is actually a kind of "indexing" database, based on a columnar store. The executor also requires refined design to fuse these hybrid search approaches effectively.
Yes, I agree. It seems each Chinese company has a secret relationship with the Chinese government.
But what they didn't notice is that the American embassy would censor your social account and content when you apply for Visa. American is the biggest surveillance country but takes it for granted.
"That China has everyone's face 3D scanned by now (multiple photos from multiple angles), everyone's voiceprint, everyone's IP, transcripts of what was said and by who"
As a native Chinese, I haven't been scanned by 3D scanner. How do you get this conclusion?
Presumably they mean that if you use your web cam with zoom they would be able to construct a 3D model of your face from the camera feed as you move your head around.
Nonsense!
If a deal hasn't been made, each side could change their mind, right? If we want to make a deal and I take great advantages on you, is it right for me to accuse you if you regrete before the deal is made?
"because Chinese developers tend to "take take take" and not contribute anything back, ever."
It's not ture. For example, Siggraph is the best conference in the computer graphics field. In 2018, the number of papers that contain Chinese or ethnic Chinese is 45%.
Besides, many Chines and ethnic Chinese reasearchers are active in many frontier technology filed such as AI and Computer Vision field.
The acadimic and industry status and constribution of Chines reashers/company are increasing rapidlly.
> the number of papers that contain Chinese or ethnic Chinese is
Nobody cares about any "ethnic Chinese" fraction. We're discussing a trade war between US and PRC, not some racial struggle between Chinese and Caucasians or whatever.
my Chinese friends feel all very similarly. While I don't agree with them, it's not productive to be insensitive to their concerns. That's not how you solve conflict.
It doesn't even make any sense. Do they think America is similarly engaged in a trade war with our own ethnic Chinese citizens? One of our most strident anti-China hawks is named Gordon Chang. And we aren't in any struggle against Taiwan.
No, I don't have to be sensitive to such delusions.
Yeah - US is a free market and should everyone in. But China should exclude whoever they like. And it is all totally fair right? Why not, because US is a free market and how dare they didn't let Huawei in?
The sconed part "Game Speed dependent on Variable FPS". I'm confused about the parameter of function "update_game()". From the code, I think the time length of "curr_frame_tick - prev_frame_tick" would increase at each timestep beacause of function "display_game()".
Every loop, prev_frame_tick is assigned the old value of curr_frame_tick before curr_frame_tick is updated. That means the value of their difference is the time in ticks it took for the last loop to execute. That difference is used to scale things like physics.
As an example, if we want to move something on screen at a constant rate in pixels per second, the number of pixels to move per update (ie frame) at 60 FPS is half as far as when updating at 30 FPS. The result is that the position at any time is the same regardless of how quickly the computer ran the simulation.
Fair enough. I think I didn't understand the difference between game time and real time before. I though the running time of "display_game()" was the game time (curr_frame_tick - prev_frame_tick). Thanks a lot for your detailed explanation.
