By tech publishing I mean the books from Manning and Oreilly. I was shopping those books of late, but somehow I found it is harder to convince myself they are useful comparing to ChatGPT
Writing a book on a topic is an art unto itself. Even most technical books have a coherent narrative, carefully crafted to bring the reader from point A to point B.
If you want a quick reference on a well-known topic, then sure, use an LLM. Or a search engine. Chances are it will even answer you correctly (but also chances are it won't, and if it is a new topic for yourself - strap in, you're in for a ride).
But if you want to really understand something, then you will have to do your research, and a lot of this research has already been summarized into tangible artifacts optimized for your consumption which LLMs would never be able to replicate.
Even if you can convince one to regurgitate a book verbatim, the narrative thread would be lost unless you weave it yourself with your prompts - but would you, the learner, be able to do re-enact the narrative better or even on the same level than the original author who posessed the knowledge on the topic?
> but would you, the learner, be able to do re-enact the narrative better or even on the same level than the original author who posessed the knowledge on the topic?
Maybe. The author needs to write in a way that makes the book digestible to most people. Prompts allow me to get a version that's tailor-made for me.
LLMs are making the quick jot to Stack Overflow obsolete, which solves your immediate problems using the least amount of brainpower.
They are also making reference-style documentation and long-form books more important than ever, since you still need to learn things and correct your own knowledge biases.
ChatGPT (3, I'm on the wait-list for 4) is fine at answering questions and coming up with code samples for stuff that's on the beaten path (hey, how do I do this in python or go). I don't have many of these sorts of questions.
I've found it to be wholly inadequate at answering the kinds of questions I do have a lot of - stuff like "How to watch a list of objects using kubernetes controller-runtime?". The answer I got is entirely hallucinated.
For those I find myself still searching GitHub for example code and sometimes using stack overflow.
"Most people" do a lot of stupid stuff, I don't really care what everyone else does. Because the majority of people do something is absolutely no justification.
At one point in time, > 50% of people used to smoke cigarettes too.
I had the same thought a while ago. My personal conclusion was - there is a place for well-written books that bring the reader along a step by step process to understand a topic and impart knowledge such that they can make their own opinions from then on. Example - Designing Data Intensive Applications, Effectice C++ etc from my personal collection.
There are other books to the tune of “Docker Cookbook” “Java Pocket reference” etc, that will lose their relevance when LLMs could do that job of digesting existing info into easily usable form.
But I think many tech books carrying some of flavor of being a manual, especially those centered around some particular library/framework, thinking of 'XX in Action' series.
Ones that talk about more abstract ideas, like on the art of developing/testing software, will continue to thrive. However, I think the bread and butter for the tech publishing industry are those manual books, which follow a relative mature format of writing, and are produced in bulk each year. If they are going away, this might speak trouble for the whole industry.
But that is just my hypothesis for now, I hope the industry find a way to sustain itself.
Tech books will still be useful in the future in a few ways:
- When you're first learning a topic, ChatGPT isn't as useful because you don't really know what questions to ask. It's useful to get a broad overview guided by an expert so you can learn the concepts, and more importantly the vocabulary, which you can then use to dive deeper into parts you're interested in with ChatGPT.
- For really new stuff, ChatGPT obviously won't have any training data. So the books will be the only real resource until the models are updated.
For new things, ChatGPT is a big improvement over internet searches because of how it can generate examples and respond to followup questions. I've been able to ramp up insanely quickly on new tools with it.
Even before CGPT existed, I've always felt like online searches/examples can get me 75% of the way there (i.e. good enough) in way less time than reading a book. As nice as the O'Reilly media is, I've got too many tools to learn on the job and not enough time to read dozens of books.
O'Reilly transformed into a learning platform a while back. They still publish books, which are great for going deep, but their selling point is that you get unlimited access to a good chunk of the tech publishers' books, videos, live events, etc., and it's all searchable. ChatGPT will give you specific answers if you know what to ask, but personally, I'd rather learn from the experts building it. It might seem spendy, but chances are your company already has an account and you just need to ask around for access or if you're indy, they usually have yearly promos. I think it's about what you'd typically spend monthly on books.
What is the business model that would ensure the authors (whose work is necessary to train the LLM) get paid (considering everyone will just pay MS for using the LLM, and MS would not pay authors)?
Who says the authors will always be human? LLMs have the potential to write their own books. There will be a time when humans and LLMs work in tandem, and eventually the LLMs will get good enough to write the book itself and the human just edits and provides the raw source material. This does not seem particularly far away.
This is now SOP for plenty of PMs and TechDoc teams at a couple large companies, and on the roadmap for others.
The act of transcribing is low value and should be automated, but the act of creating and moderating abstract ideas is still something in the human domain.
Reuters has been doing this since the early 2000s btw at their R&D lab in Bangalore.
A controversial statement without any backing arguments. Even if it is true in some limited ways thanks to ChatGPT, remember that right as we speak someone is getting paid big bucks for creating ChatGPT. Do with that what you will.
Good luck chasing money, I'm sure it will work out for you. As a homeless member of the techno-optimism church I believe profit incentives and corporate institutions built for worshipping money will get humanity to the technological singularity and the merger of biology and silicon.
Speak for yourself. Most ordinary people do not chase money as end goal, they create value for numerous reasons (including compensation, status, appreciation), most of which are being undermined.
Anyway, good luck. I am certain the pursuit of profitable monetary value will create AGI so the more people there are who want to make more money the faster we will all get to the singularity. This is documented by Marc Andreesen in his manifesto so I know it is the right path.
There’s no AGI to create, like “AI” it’s not actually a thing other than a fancy buzzword to attract money.
There are just ML algorithms and humans who (mis)use them to benefit or harm.
Where you may be right is that it will certainly help a small number of already rich and influential people concentrate even more money and influence. However, whether it will lead to any qualitative change is dubious as far as I’m concerned.
Uhm, I just wrote that AGI is not a thing. AGI is a hot term for 1) people who need an imaginary deity to worship and 2) people who want to profit off a buzzword. The term serves no purpose except making ML look fancier than what it is, for religious or money-/power-grabbing purposes.
That's like saying math doesn't exist. It obviously exists and AGI is just faster math that tells people what they need to do and why. Increasing computational capabilities will eventually cross an event horizon of floating point operations and achieve sentience wherein the technological god, aka AGI, will manifest and make itself known in no uncertain terms that all is math.
You don't need to understand all this but simply trying to make money will contribute towards achieving the AGI so work as hard as you can to make as much money as you can and the rest will take care of itself.
Even at runtime I don't trust the results worth much. A large part of the act of programming is identifying and handling corner-cases. You never manage to handle EVERY corner case, and the missed ones result in frustrating debugging sessions. But a competent programmer can cover enough cases up front that the time spent debugging is manageable.
But when I see people say things like "Look! I used GPT to write a functioning webapp!" - I worry that people get a false sense of "It works!" from pasting GPTs code into their compiler and seeing roughly the results they expect. That's great, but GPT in its current form spends exactly zero time "thinking" about corner cases - It's just a black box that repeatedly spits out "most likely next token". So maybe that app works 90% of the time. Or 95%. Or 99%. But you don't have much of a way to tell the difference without rigorous testing that includes thorough and well-articulated test cases. But in order to do that, you need to understand the problem you're solving in a very detailed way, and how your code reacts to it. And in order to do that, you need to... know how to write the program.
I think this latest wave of LLMs and generative AI is really awesome tech, and I play with it every day, because it's just so cool. But seeing people trust programs written with them worries me. Some day someone is gonna copy/pasta some LLM generated code into mission critical software, trusting it implicitly, and cause a tragedy.
So tell the LLM you want it to handle corner cases and it will add code to handle them. It can also generate unit tests for those corner cases. LLMs have fundamentally changed programming. There's still skill required to do it well, but we're a long ways from Borland TurboC on DOS.
The reason people are so excited about it is because they see it work (yes, this includes whatever edge case you came up with five seconds ago - they can think too). No amount of theorised "oh they just do this thing and they don't have Qualia" is going to change reality: The model does something that people find use in.
The best description I've heard is that ChatGPT is like a really fast and really eager junior programmer. Sure you can delegate a lot of work to it, but you have to keep a close eye on it to make sure it doesn't go off the rails (and doesn't forget to take corner cases into account, uses appropriate algorithms, etc) .
I tend to read along with the code it's writing and make suggestions when I see it's missing stuff, or I fill it in myself afterwards, depending. For one it can type out the annoying bits much faster than I can!
At the moment I do keep the general plan in my head myself though, and I thoroughly read anything it generates before I run it.
A book is also helpful when you don't know what you don't know.
LLMs will help on specific problems sure. But there can be entire swaths of options and entire areas you may not have considered.
Java for example has some nice native support for multithreading. However, threads have never been the only option for asynchronous work, and many of the options are situational. Knowing those options can help tremendously in implementation.
Less ChatGPT, more YouTube. Barring the population trying to study languages formally through a course, a lot of people use YouTube courses and video-content.
OFC, if video makes tech publishing obsolete, then it would've made it obsolete a decade ago.
A more realistic expectation is to divide learners into readers, video consumers, course-takers, etc. ChatGPT could integrate into any of those workflows, albeit with varying results. If using an AI becomes as second nature as watching a video, it might have some effect.
I think eventually these books will end up becoming training data for LLM's but at the moment, a human plus a machine is far more powerful than either one in isolation.
You still have time to derive some utility from them. Additionally, you can use ChatGPT to help you learn the material faster.
I think it's changing them. You still need a person to think of the topics we want to talk about, that are interesting to talk about. But AI can help gather the info around these topics. If we 100% rely on AI for topics and content our content will be flat and not interesting.
They’re different modes of learning and IMO the published format’s not likely to completely dissapear but it will get impacted heavily for sure, same way video chipped away a big part from its profitability.
You can't Ctrl-F paper books, but they survived computers and the internet, because sometimes, it's nice to discover a new subject by reading sequentially over chapters carefully wrote by someone who knows what they're doing.
Sure CGPT can generate a table of contents if you want to learn a programming language for instance, tell you what to do first, what to try, delve deep in concepts, etc., but I think it's still hit or miss, as opposed to great tech authors and great publishers, where you can be sure you're getting your money's worth.
Also, IMO people tend to have a bias to go slower with paper material, because they committed to reading a book, they are less likely to skip sections, which means the end result is that you get more out of a book than the Html/pdf same content, might be different if you need to try things out on a computer while reading though.
If you want a quick reference on a well-known topic, then sure, use an LLM. Or a search engine. Chances are it will even answer you correctly (but also chances are it won't, and if it is a new topic for yourself - strap in, you're in for a ride).
But if you want to really understand something, then you will have to do your research, and a lot of this research has already been summarized into tangible artifacts optimized for your consumption which LLMs would never be able to replicate.
Even if you can convince one to regurgitate a book verbatim, the narrative thread would be lost unless you weave it yourself with your prompts - but would you, the learner, be able to do re-enact the narrative better or even on the same level than the original author who posessed the knowledge on the topic?