3 ms·

> "n AI it's been a cliche now how many introductory blog posts and into YouTube videos and "how do I start"" one of the frustrating things its really hard to

> "n AI it's been a cliche now how many introductory blog posts and into YouTube videos and "how do I start"" one of the frustrating things its really hard to find great advanced content, and when you reach out to HN readers, they are like 'read really advanced PHD papers'. thats not what I mean at all >.<

bonoboTP 6y ago
There are textbooks, but people (myself included sometimes) want something flashier something more palatable. But other than conference talks and papers, textbooks and lecture slides, there just isn't much out there. It's a consequence of how the sausage is made. Textbooks take years to write so only established topics can be included in them. Researchers are under time pressure and chase performance criteria. Publishing at a good conference pushes you along your PhD path, but writing a blog post for a niche advanced target audience of maybe 100 readers is not always worth it. Especially when you have other duties beyond research like teaching courses. Industry blog posts are good, but also cannot go very deep as the audience would be lost. Also the nature of advanced stuff is that there is less of a clear established path forward. There are tons of small specialized communities that often don't know much about each other, they are in different departments etc. Your best bet in becoming more advanced is to use the above mentioned resources and work with people who are more advanced than you, either in academia or industry. There is no law of nature that everything must be achievable by browsing the free internet without leaving the house. An important lesson when you get to a point in your studies is that there is no set Platonic chapter-by-chapter structure in knowledge and science. A common failure mode is to think that learning is about leveling up in some achievement tree like in duolingo, and being puzzled where the next chapter is. You have to seek it out, tackle a practical problem either from your own idea and doing a side project or by joining a group as an assistant or a company as a junior engineer/data scientist. A lot of it is "dark knowledge", hidden inside organizations, people learning from one another. There are best practices that "everyone" just knows to do by discussing by the coffee machine, it's not in any book or paper. And papers are written in an obtuse language to not give away too much (because that will be in the next paper). The interests of an advanced practitioner reader are markedly different from the paper writer researcher's incentives, so this can make papers hard to read and decipher what to take away from them. The goal of a paper is to frame a small incremental tweak in the context of the larger scale literature in such a way that the reviewers are sufficiently impressed to click accept. Papers are obsessed with how they fit in the academic community's pursuit. They aren't tutorials or howtos or guides. Fortunately people are now releasing more and more code. The code is often more enlightening than the aggrandized math of the paper (which often boils down to few lines of code).

bonoboTP 6y ago

There are textbooks, but people (myself included sometimes) want something flashier something more palatable. But other than conference talks and papers, textbooks and lecture slides, there just isn't much out there. It's a consequence of how the sausage is made. Textbooks take years to write so only established topics can be included in them. Researchers are under time pressure and chase performance criteria. Publishing at a good conference pushes you along your PhD path, but writing a blog post for a niche advanced target audience of maybe 100 readers is not always worth it. Especially when you have other duties beyond research like teaching courses. Industry blog posts are good, but also cannot go very deep as the audience would be lost. Also the nature of advanced stuff is that there is less of a clear established path forward. There are tons of small specialized communities that often don't know much about each other, they are in different departments etc. Your best bet in becoming more advanced is to use the above mentioned resources and work with people who are more advanced than you, either in academia or industry. There is no law of nature that everything must be achievable by browsing the free internet without leaving the house. An important lesson when you get to a point in your studies is that there is no set Platonic chapter-by-chapter structure in knowledge and science. A common failure mode is to think that learning is about leveling up in some achievement tree like in duolingo, and being puzzled where the next chapter is. You have to seek it out, tackle a practical problem either from your own idea and doing a side project or by joining a group as an assistant or a company as a junior engineer/data scientist. A lot of it is "dark knowledge", hidden inside organizations, people learning from one another. There are best practices that "everyone" just knows to do by discussing by the coffee machine, it's not in any book or paper. And papers are written in an obtuse language to not give away too much (because that will be in the next paper). The interests of an advanced practitioner reader are markedly different from the paper writer researcher's incentives, so this can make papers hard to read and decipher what to take away from them. The goal of a paper is to frame a small incremental tweak in the context of the larger scale literature in such a way that the reviewers are sufficiently impressed to click accept. Papers are obsessed with how they fit in the academic community's pursuit. They aren't tutorials or howtos or guides. Fortunately people are now releasing more and more code. The code is often more enlightening than the aggrandized math of the paper (which often boils down to few lines of code).