casualscience

6 ms·

31.
by casualscience 1y ago
Just went through that thread, I can't believe this wasn't a team of like 20 people. It's crazy to me that apple would put one guy on a project this important. At my company (another faang), I would have the ceo asking me for
32.
by casualscience 1y ago
How can you create a 'pressure differential' without deflecting some of the air away? At the end of the day, if the aircraft is moving up, it needs to be throwing something down to counteract gravity. If there is some pressure dif
33.
by casualscience 1y ago
How can you create a pocket of 'lower pressure' without deflecting some of the air away? At the end of the day, if the aircraft is moving up, it needs to be throwing something down to counteract gravity.
34.
by casualscience 1y ago
This sounds like a smart comment, but the main reason you shouldn't take in vitro studies as indicative of real medical outcomes is largely due to unknown bio availability when consuming realistic doses. However, this study shows that
35.
by casualscience 1y ago
blood clots: https://www.nature.com/articles/s41591-023-02223-9 . Evidence in vitro suggests enhanced platelet activity. Plasma levels of erythritol are sustained for >2d above thresholds associated with platelet hyp
36.
by casualscience 1y ago
yeah, I used to work in the medical tech space, they love to tell you how much you should be in it for the mission and that's why your pay is 1/3 what you could make at FAANG... of course, when it came to our sick customers, they
37.
by casualscience 1y ago
You need to setup the problem for the llm. If I am getting an error, I can normally piece together the 10k lines of relevant code much more quickly than I can track down the bug. Llms are still a big speed boost there
38.
by casualscience 1y ago
probably not? MS deploys those models themselves, they don't go to OAI at all
39.
by casualscience 1y ago
95 is a pretty random cutoff. But you can see for yourself the number of 65%, 75%,85%,90% positively reviewed games has increased similarly: https://steamdb.info/instantsearch/?refinementList%5Btags%5D...
40.
by casualscience 1y ago
source?
41.
by casualscience 1y ago
In big companies, this is a bit slower due to the need to migrate entrenched systems and org charts into newer workflows, but I think you are seeing more productivity there too. Where this is much more obvious is in indie games and software
42.
Diffusion Language Models in 2 Minutes [video](youtube.com)
1 points by casualscience 2y ago|0 comments
43.
by casualscience 2y ago
The story goes that the US subsidizes all the drug prices around the world. Without the insane US margins, drug companies wouldn't ever do the needed R&D... so they say.
44.
by casualscience 2y ago
> Multi-backend is doomed Finally! Someone says it! This is why the C programming language will never have wide adoption. /s
45.
by casualscience 2y ago
While I agree with your general sentiment, this is not a great study. They did not control for weight, physical activity, drug use (other than tobacco and alcohol), socioeconomic status, etc.... When you take people who have substance abuse
46.
by casualscience 2y ago
> Notebooks are often used as a playground to easily have access to an accelerator. In this case, the execution environment is often set up differently and with different dependencies. In this case, you make your code work in a colab, sm
47.
by casualscience 2y ago
The best thing about root was how it handled data loading. TTree's, with their column based slicing on disk, are such a good idea. Ever since I graduated and moved into industry, I've been looking for something that works the same
48.
by casualscience 2y ago
I'm confused as to the use cases for this? Are you saying if I want to fit some "magic numbers" in my cpp program, I can now do that by pulling in discograd and wrapping those numbers with some code that says "please fit
49.
by casualscience 2y ago
why can't I start my day at noon?
50.
by casualscience 3y ago
Oh nice, I made a video covering the history here (as well as some of the earlier scientific progress on the existence of particles): https://www.youtube.com/watch?v=_uDT2lNi5bQ&ab_channel=Casua... Can't miss a cha
51.
by casualscience 3y ago
That literally requires human annotators. If you can do even 75% as well as a human but at 1000x the scale, you still win using the LLMs. Not to mention, I personally do think on a simple task like "Does document A support response B&q
52.
by casualscience 3y ago
hmm, I think the entire FIRE movement is basically about this, no? Some people want to just do gardening or whatever, but if your hobby is science, that's totally fine too. Not every kind of science can be done on the cheap, but basica
53.
by casualscience 3y ago
I did so after college, had no money, parents were immigrants with no money but always told me to "follow my dreams". PhD payed enough to eat and live a decent life as a 20-something with no dependents. As I got older, I realized
54.
by casualscience 3y ago
Much of the foundational science we learn about in the history books was done by bored rich people. Becoming a bored rich person is still a great strategy if you want to do foundational research. The risks you need to take in order to do th
55.
by casualscience 3y ago
Yes exactly. I think it's an interesting case. Where/how do you legislate this? Is it based on intent (you literally are training your model to be cooperative), or is it based on effect (your company is having this effect regardle
56.
by casualscience 3y ago
My understanding is that further than just using software, the software is optimizing to increase maximum profits for the industry by causing a cartel block of people who won't lower prices. Any individual can go out there and build a
57.
by casualscience 3y ago
> the kind of world-ending AI we're worried about could be created and/or released by accident by a single group or individual, could self-replicate on the Internet, and would be unstoppable once released. I also worry every t
58.
by casualscience 3y ago
An attention layer transforms word vectors by adding information from the other words in the sequence. The amount of information added from each neighboring word is regulated by a weight called the "attention weight". If the atten
59.
by casualscience 3y ago
That could work, I think denoising a song to be a perfect match to the original recording is probably a very hard problem, so hard that your model will still need to be robust to some deviation from the original track, and therefore you nee
60.
by casualscience 3y ago
I'm an MLE, I would probably chop the songs into short segments, add noise (particularly trying to layer in people talking, room noise, and apply frequency-based filtering), and create a dataset like that. Then I would create contrasti

More ›