raindear

10 ms·

1.
by raindear 10mo ago
There has been a slow progress in computer vision in the last ~5 years. We are still not close to human performance. This is in contrast to language understanding which has been solved - LLMs understand text on a human level (even if they h
2.
by raindear 11mo ago
Are dead ReLUs still a pronlem today? Why not?
3.
by raindear 1y ago
How far can you go by improving the curriculum? Start simple. Find a shorter and shorter sequence of examples that gives you thd best result. What is the shortest sequence to get to some perplexity? Why?
4.
Agenda 47(en.wikipedia.org)
7 points by raindear 1y ago|0 comments
5.
by raindear 2y ago
Progress is what happens thanks to AI skeptics busy defining model limitations. The limitations set attractive bars to pass.
6.
Astronomers Are Keeping an Eye on This Asteroid's Odds of Hitting Earth(nytimes.com)
3 points by raindear 2y ago|1 comments
7.
by raindear 2y ago
I read that deepseek was trained on western llm output. So it is expected to have the same biases.
8.
by raindear 2y ago
NYT: https://news.ycombinator.com/item?id=42403886
9.
Creating 'Mirror Life' Could Be Disastrous, Scientists Warn(scientificamerican.com)
1 points by raindear 2y ago|2 comments
10.
Is rooftop solar a fatal flaw in the Coalition's grand nuclear plans?(abc.net.au)
3 points by raindear 2y ago|1 comments
11.
Ask HN: Finding a part time dev job
6 points by raindear 2y ago|2 comments
12.
Twitter now returns "JavaScript Is Not Available"(twitter.com)
2 points by raindear 2y ago|0 comments
13.
by raindear 2y ago
Molecular Biology research is much more funded than computer science. At least in Academia.
14.
by raindear 2y ago
Yes, cDNA is a DNA fragment that does not interact with the patient's DNA in this treatment. Wikipedia: "Once amplified, the sequence can be cut at each end with nucleases and inserted into one of many small circular DNA sequence
15.
The Illusion of State(twitter.com)
1 points by raindear 2y ago|0 comments
16.
by raindear 2y ago
But why do transformers perform better than older language models including other neural language models.
17.
by raindear 3y ago
There are around 40k Hamas gunmen. So the reported numbers can get that high.
18.
by raindear 3y ago
What would you do if an enemy army crossed the border and slaughtered 1000 men, women, children, party goers and elderly?
19.
by raindear 3y ago
Israel withdrew from Gaza in 2005. There have not been tanks there since.
20.
by raindear 3y ago
Gaza Strip borders with Egypt too.
21.
by raindear 3y ago
The article sympathizes with only with one side of a horrible war, and only about losing its internet connection. It's surrealistic. It's like reading an article about Berlin citizens losing telephone connection in May 1945.
22.
by raindear 3y ago
For example, you can create an embedding for an image using a neural net that has been trained to receive an image and output a vector of 1024 floats, which represent the content of the image. This vector is a lossy compressed version of th
23.
by raindear 3y ago
A competition to self driving cars? If urban flight is possible in 10 years, will people need self driving cars?
24.
by raindear 3y ago
It's not available for smartphones.
25.
Intelligence Research Resource Task Force Releases Final Report(whitehouse.gov)
2 points by raindear 4y ago|0 comments
26.
Ask HN: When to Stop Negotiating?
1 points by raindear 4y ago|0 comments
27.
by raindear 4y ago
Yes, the training is based on predicting a missing word. At first, such a task allowed GPT2 to generate paragraphs that were surprisingly coherent, at the user's request. ChatGPT added a novel ability of conversation. Regarding timesta
28.
by raindear 4y ago
The date of the post is April 1st.
29.
by raindear 4y ago
It's not the task, it's the co-workers.
30.
by raindear 4y ago
But it could be detected with RF.

More ›