isaacfung

8 ms·

31.
by isaacfung 3y ago
A lot of polynomial time algorithms for combinatorial optimization problems can be interpreted as primal dual algorithms for the corresponding LPs, e.g. mst, matching(bipartite or general graph), network flow, matroid intersection, submodul
32.
by isaacfung 3y ago
Yea, Daniel Spielamn and Shang-Hua Teng won the Gödel Prize for their work on smoothed analysis of simplex algorithms. They introduced a way to formally study the worst case complexity of algorithms when the inputs are randomly perturbed by
33.
by isaacfung 3y ago
First of all I dont think we should aim at 100% precision and 100% recall. Instead it is more realistic to use the result as a filter for more downstream testings and treat it as a mean to increase the cost of cheating. Also we can use more
34.
by isaacfung 3y ago
Are You Normal? Hint: No https://allendowney.github.io/ProbablyOverthinkingIt/gaussia...
35.
by isaacfung 3y ago
https://www.reddit.com/r/ChatGPT/comments/18qhcst/yooo_gpt4_...
36.
by isaacfung 3y ago
Something like this? https://imgur.com/a/hPAaZUv https://huggingface.co/spaces/Qwen/Qwen-VL-Plus You can also ask it to give you bounding boxes of objects.
37.
by isaacfung 3y ago
Wow, the other link is so much faster. Almost the same speed as the much weaker llava demo on huggingface.
38.
by isaacfung 3y ago
The examples look very promising but the queue time of the huggingface demo is a bit long I actually got a timeout error :(
39.
by isaacfung 3y ago
This would never happen but in an ideal world, we should be able to click on a citation to jump to the part of the paper that is being referenced and each paper page should have a discussion board so we can easily communicate with the autho
40.
by isaacfung 3y ago
What device and app are you using to read the document?
41.
by isaacfung 3y ago
This, betterexplained, ritvikmath, SeeingTheory will give you a very solid math background(I think they are better than 90% of the intro math classes in colleges).
42.
by isaacfung 3y ago
I think that is partly why LLMs are bad at math and often fail at counting subsequences. Play with the tokenizer and you see long numbers are split into groups of 2 or 3 numbers. https://huggingface.co/spaces/Xenova&#x2
43.
by isaacfung 3y ago
Who else was hoping to see a real MRI scanner built with LEGO?
44.
by isaacfung 3y ago
Whisper, Stable Diffusion, VoiceBox, GPT4 vision, DALL.E3 Other breakthroughs in graph machine learning https://towardsdatascience.com/graph-ml-in-2023-the-state-of...
45.
by isaacfung 3y ago
Depends on what you mean by search. Do you consider all Question Answering as search? Some questions require multi-hop reasoning or have to be decomposed into simpler subproblems. When you google a question, often the answer is not triviall
46.
by isaacfung 3y ago
If they really want to catch these people, you wont be seeing only a few examples over half a decade. As mentioned in the article, even Hu Xijin (the former editor-in-chief and party secretary of the conservative popular media Global Times)
47.
by isaacfung 3y ago
You tried to prove these are not independent moves by providing a few incidents out of a 1.4 billion country over a span of several years. One comment in the first article you linked said 99% of software engineers in china use github. The C
48.
by isaacfung 3y ago
I know. I live in Hong Kong. Some of these are worrying but a few of them seem questionable. From what I can find, the previous asian games opening also wasn't attended by many world leaders. It's not Olympic. https://e
49.
by isaacfung 3y ago
Can you provide some sources about the flood destroying military forces and Hangzhou government raiding temples? About Huawei stores being empty, some seems pretty crowded at least on the first day. https://youtu.be/YFeE2xHQ
50.
by isaacfung 3y ago
Some people see innovative ways to combine art and technology. Some people see money.
51.
by isaacfung 3y ago
These may be related. https://www.reddit.com/r/datascience/comments/11vzjhi/statis... https://www.reddit.com/user/eamonnkeogh/submitted/?sort=top
52.
by isaacfung 3y ago
Dont send the cc info. Use private key to authenticate yourself. Digitally sign the transaction id and the date so people cant record the info and reuse it. I guess the issue is that people may lose or forget their keys or people can gain a
53.
by isaacfung 3y ago
I am an idiot. I just read the comments without looking at the linked page.
54.
by isaacfung 3y ago
OpenAI used this approach for CLIP before blip and llava. CLIP is used to encode the text prompt in stable diffusion. Not sure about Dalle.
55.
by isaacfung 3y ago
It was all over hackernews. Just google "hacker news gpt4 mixture of experts".
56.
by isaacfung 3y ago
Two tower isn't a new term afaik. It's often used in the context of recommendation systems where the two modalities are users/query and items. https://github.com/creyesp/Awesome-recsys
57.
by isaacfung 3y ago
https://github.com/WooooDyy/LLM-Agent-Paper-List#114-reasoni...
58.
by isaacfung 3y ago
https://github.com/WooooDyy/LLM-Agent-Paper-List
59.
by isaacfung 3y ago
LLM has no long term memory(limited by context size or you have to do fine tuning which isnt efficient) and can't interact with the outside world. You can augment it with retrieval systems(vector store, graph db, sql db + lexical/
60.
by isaacfung 3y ago
LIMA: Less Is More for Alignment https://arxiv.org/abs/2305.11206 AlpaGasus: Training A Better Alpaca with Fewer Data https://arxiv.org/abs/2307.08701 Textbooks Are All You Need II: phi-1.5 tec

More ›