People Who Should Learn to use ChatGPT
You are probably one of them; #ChatGPT in 2023 is poised the way the World Wide Web was in 1993
My views on ChatGPT are evolving. Here are my current thoughts.
Think of ChatGPT as a tool that a lot of professionals should learn how to use. Probably not as many people as need to learn email, but probably more people than need to learn JavaScript Object Notation (much as I love JSON). Roughly comparable to Excel.
As with Excel, ChatGPT can have a variety of uses, but taking advantage of it may require some upfront learning. With optimal prompting, ChatGPT can perform some tasks surprisingly well. But with sub-optimal prompting, the results can be useless.
Also, some Excel users are prone to sinking time into it trying to get it to perform tasks that are more appropriately handled by other software tools. As with Excel, knowing what ChatGPT should and should not be used for will not be immediately obvious to a novice.
This essay is not meant to be anything like an introduction to ChatGPT. In a sense, it is too early to do that. Most of the potential uses of the software have not been imagined, much less tried and tested.
I am just trying to encourage you to become an early adopter. If you are one of the first people in your organization, or one of the first people in your profession, to learn how to form prompts well and how to incorporate ChatGPT’s output into your work, then you will probably enjoy a productivity advantage for several years. If I am correct, then the laggards will lose out to the early adopters. You are better off eating than getting eaten.
What will reduce the advantage of early adopters will be the emergence of a training infrastructure around ChatGPT. That suggests that one can earn a nice living as part of that infrastructure.
Be wary of people who anthropomorphize ChatGPT. It is not a high-skilled immigrant coming for your job or a prodigal child soon to become the ultimate genius. It is software, with intrinsic limitations.
the fundamental idea of a generative-language-based AI system like ChatGPT just isn’t a good fit in situations where there are structured computational things to do. Put another way, it’d take “fixing” an almost infinite number of “bugs” to patch up what even an almost-infinitesimal corner of Wolfram|Alpha can achieve in its structured way.
Thanks to a reader for pointing me to Wolfram’s essay.
between the human-like nature of chat and the fact that written material is harder to immediately evaluate, many people tend to avoid explicit prompt construction in ChatGPT. But that is a mistake! More elaborate and specific prompts work better.
When you search Google, you may have to formulate your query carefully to get it to home in on what you want. ChatGPT is even more tricky to use.
Much like students learn Excel and calculators - and even how to use more advanced formulas and engineering calculators - how do we outfit students to make sure AI is working for them (and not the other way around)? Also, related to the next forecast, how do we equip the next generation to know what they can trust and how to evolve their own mental judgment as AI spits out answers?
He also writes,
I hope this new tech evolves education to be more about learning how to think. How to find answers. How to connect dots. How to express yourself creatively (and stand out, merchandise your ideas, galvanize support for unpopular views).
I asked ChatGPT to tell me how to play “Can’t Buy Me Love” on the guitar. It responded with one of it’s “hallucinations,” (an anthropomorphism that people have come up with to describe ChatGPT’s incorrect outputs that the user has to spot for himself.) It gave me an alignment of chords with lyrics that was clearly wrong.
YouTube is clearly better. ChatGPT ended up pointed me to a YouTube video, but that was another hallucination: when I clicked on the link, YouTube said that the video had been taken down. To the extent that most children want to learn “how to do” rather than investigate ideas, my guess is that if you want to turn those kids loose to learn on their own you’re better off pointing them to YouTube than to ChatGPT.
A machine-learning algorithm can get better at doing things by giving it more data and/or more feedback, but that learning process does not necessarily progress the way a human’s will. If ChatGPT can’t give a correct answer to a calculus problem, then feeding it more data is not going to help it “learn” calculus.
Wolfram writes,
given [ChatGPT’s] dramatic—and unexpected—success, one might think that if one could just go on and “train a big enough network” one would be able to do absolutely anything with it. But it won’t work that way. Fundamental facts about computation—and notably the concept of computational irreducibility—make it clear it ultimately can’t.
Even with its limitations, ChatGPT and its relatives probably will have many valid use cases. But I think that it is premature to talk about jobs that will be replaced by ChatGPT, just as it was premature a few years ago to talk about jobs that will be replaced by self-driving cars.
The problem with self-driving cars is that the algorithms sometimes make mistakes that no decent human driver would make, so you need the ability for a human to take over. ChatGPT is in that state currently.
That said, I think that the use case for self-driving cars is much better right now than most people are willing to acknowledge. There are many reckless drivers who make mistakes that no self-driving car would make, and those reckless drivers cause most of the bad collisions. I would love to see a feature that allowed a self-driving car to take control away from such humans—the guy who drives drunk or the guy who goes 50 percent over the speed limit, for example. I’m not sure if we can agree on what the equivalent is of a reckless driver where we would want ChatGPT to take over for humans.
Substacks referenced in this essay:
@
@
I don't think describing ChatGPT as 'software' creates useful associations. It is software only in the most superficial sense: it is not hardware but runs on hardware. Excel is software: basically a big bunch of step-by-step instructions written by humans, executed by ultra-moronic bean-counting homunculi. This means it is possible for users to form accurate mental models of at least parts of Excel. This is not true for ChatGPT and similar entities. I feel it is much more illuminating to think of them as evolved artificial life forms, or perhaps communities (synusia) of such life forms. Training sets and procedures are their environment, and space within the neural network's parameter set is the resource they compete for. They behave, like animals do, rather than function, as machines and algorithms do. People who train these networks are closer to farmers practising artificial selection to create new breeds than to programmers planning and creating step-by-step instructions in their minds and exporting them in forms executable by ultra-moronic bean-counting homunculi. Very old, very complicated, and badly managed software _can_ reach a state when it behaves rather than functions, but software developers recognize this as an evil and given opportunity work hard to remedy it. Users who haven't formed accurate mental models of the relevant parts of software they are interacting with can also feel that the software behaves rather than functions, but this is a consequence of their having inaccurate mental models when accurate mental models exist (the software itself being such a model). ChatGPT and its peers aren't such models and therefore forming mental models of their behavior is like forming mental models of the behavior of your coworkers, or of a dog. A number of academic papers have been published already investigating how this or that large neural network does what it does. I looked over a couple and they feel like psychology or physiology papers picking apart rats rather than like computer science papers picking apart algorithms. In fact, the very idea of investigating how an algorithm does what it does is almost a contradiction in terms, because the algorithm is 'how' boiled down to essence. We have now arrived at the point where we can appreciate Searle's Chinese Room gedankenexperiment. It is now obvious that Moldbug was right when he argued 15 years ago that it is the Room rather than the man in it who speaks Chinese, but that this does not lead into the paradoxes Searle imagined because the rules the man follows in making the Room work are not globally intelligible to him.
ETA: attentive readers will notice that there is a continuum between globally intelligible algorithms and GPT-class digital life forms, with Excel somewhere in the middle. As a very rough estimate of complexity, the source code of OpenOffice (main trunk, exclusive of tests, extras and assets such as document templates) is ~350Mb of C++, but of course it is very much redundant; 7z compresses it to 40Mb. GPT-3 has 175G parameters and they use int8, so it's 175Gb and it's a good bet that it does not compress at all. On the other end, Euclid's algorithm, which gave algorithms their name, is just dozens of bytes.
Another limitation of ChatGPT at present: I just tried to use it and it came back with the following error message at 9:30 AM on a weekday (prime time?):
"ChatGPT is at capacity right now
Get notified when we're back online"