astee

It's interesting because I use them every day all day for this now.

You have to "gut check" the answers and know when to go deeper.

A lot of answers are low stakes, and it's OK to be a little wrong if it helps go in the right direction.

If I ever need a fast matmul, you're hired.

Yes. But we're talking about children too - not just adult voters.

And the app collects every click, every face photo, all contacts, every keypress on external links, everything. The full social graph, shaping the trends of the younger generation.

Writing (parts of) an operating system, networking stack, interpreter, and text editor. It dispels all the magic.

Getting over the whole idea of beautiful or perfect code. Realizing that code is a medium of expression and that there is no single right way to do everything. Beginning to think of code as a clay-like medium rather than literary prose that needs to be beautiful like a cathedral.

Seeing how some of the most experienced and productive programmers do things in extremely simple straightforward ways, with no unnecessary cleverness, indirection, abstraction, or anything like that.

Seeing how linus torvalds is right. It's really all about the data structures. Once I understand the data structures and their relationships, I can understand any system. Dispelling the myth that some things are just "too complicated". Almost every problem or system I've ever come across can be reduced to a page of data structures.

Realizing the power of fast feedback loops. How just about any requirement or problem can be expressed as a test case. And that once you have the test case you can fix anything.

Prioritize output early.

You can't do great without slogging through mediocre. Don't be afraid to suck. Don't stop at the first failure.

And don't worry about originality. Creativity comes from doing. Experience begets ideas.

There is a lot of work where a human body + mind is cheaper and easier than automation and likely will be for the foreseeable future. We will need lots of human labor until you have embodied robots similarly adaptable to a human at <= $20/hr.

Automation puts downward pressure on wages/working conditions. But we must also consider globalization pushback and aging population pyramids that put upward pressure on wages.

Conclusion? Varies.

It doesn't replace a skilled programmer. It just turns you into an editor and curator, multiplying productivity on some tasks by 10X+.

It will give incorrect code, but you can guide it toward a correct solution by asking it to fix the problem. Normally you don't even have to say exactly what is wrong.

For example, I got it to implement a basic bittorrent tracker server in Go in about 5 minutes. I didn't even have to point it to the RFC. I just said to consult the bittorrent RFC. It gave me back a server with /announce and /scrape endpoints. I then asked it to implement the functions using a struct for the requests. It correctly deserialized the URL-encoded sha1 info hashes from the /announce endpoint on the first try. I didn't even have to mention that detail.

It can also help you explore solutions. I asked it about algorithms to learn policies for 2-player zero sum games. It gave me a description of min-max, MCTS, reinforcement learning, deep neural networks. I then asked it to describe the pros/cons of each, which it did. I asked it to show an example of a reinforcement learning algorithm in python from scratch, which it did in about 10 seconds.

grep for "incr" and add "mult" below it ¯\_(ツ)_/¯

For naturals, there is significant overlap in the training style that produces maximum strength gains and maximum size gains. Larger muscles are stronger muscles. Studies have shown very similar size gains between groups performing high weight sets and more moderate "hypertrophy" style sets, though training with higher weight maximizes strength gains. One thing not revealed in short term studies, but which I have observed[1], is that training for strength early on eventually produces better gains in size, since when you do shift to "higher rep" training, you will be able to use more weight -> bigger muscles. Regardless, I don't think it's all that important to stress over these details, and progressively overloading (increasing weight/reps) the compound movements (squats/deadlifts/bench press/overhead press) over time is far more important than rep schemes and intensity.

For steroid users, the equation changes some, as they benefit more from very high volume, high rep training that pumps the muscles full of blood. Natural lifters cannot recover quickly enough from this sort of training.

[1]standard disclaimer about taking this with a grain of salt as I'm just some internet stranger, but I do have > 10 years experience with this

This is the norm for me. If I am awake, I almost always have music playing in my head, from pop songs to classical to little nondescript melodies. Sometimes this can be quite annoying.

Hits home here. I started reading HN as a college freshman. At the time, a lot of the titles were gibberish to me. Over the next few years I sucked in articles and comments until I had (what I thought was) a pretty good mental model of the industry.

Well that model was... quite a bit off.

But the filter bubble has its merits. A little daily exposure to HN keeps the ideas churning when you're mostly doing mundane stuff.

Interesting. I wonder if the musical practice led to the mental music. I am also a musician, but I'm not certain whether the mental music began before or after I started playing (I started playing more than ten years ago as a teen).

In my case, songs or melodies are nearly always playing. If I'm awake, chances are that something is on in my mind's radio. Conversation and writing seem to be the only activities that consistently quiet the music, though there are other times during the day when it's less noticeable.

This isn't always pleasant, to be honest, and since I first noticed it a few years ago I've wondered how common it is and whether training can stop it. No luck so far.