Posts

ChatGPT is settling the Chinese Room argument 2022-12-04T20:25:48.853Z

Comments

Comment by averros on ChatGPT is settling the Chinese Room argument · 2022-12-05T01:55:26.829Z · LW · GW

The point is not "not understanding sometimes", the point is not understanding in a sense of inability to generate responses having no close analogs in the training sets.  ChatGPT is very good at finding the closest example and fitting it into output text.  What it cannot do - obviously - is to combine two things it can answer satisfactory separately and combine them into a coherent answer to a question requiring two steps (unless it has seen an analog of this two-step answer already).

This shows the complete lack of usable semantic encoding - which is the core of the original Searle's argument.

Comment by averros on ChatGPT seems overconfident to me · 2022-12-04T19:56:37.570Z · LW · GW

The questions about the wing lift show complete lack of understanding of the basic physics by ChatGPT.

The Coanda effect is caused by viscosity (the adjacent layers of air transfer some of the momentum of molecules between them), and the wing material also transfers momentum to adjacent layer of the air (causing parasitic drag).   This causes the air flow to "stick" to the top surface (to understand how just look at the streamlines to see how the speed gradient from wing to undisturbed air bends the flow) and so be deflected downwards, which creates lift (Newton's 2nd and 3rd laws).

Without viscosity there is no lift (this is known as Kutta condition).

Now. Bernoulli's principle is independent from viscosity (it's a consequence of conservation laws), and so cannot be a correct explanation of aerodynamic lift.

(The loss of "stickiness" of laminar airflow over upper wing surface by transitioning to turbulent flow is known as wing stall, and causes rather abrupt loss of lift when the angle of attack goes approximately above 18 degrees - again, according to Bernoulli principle the stalls would be impossible, but any pilot can easily demonstrate that they do, indeed, exist).

Just face it - ChatGPT is just a Chinese Room.  Repeating any popular nonsense from the training set without any deeper understanding or generalizing.

Comment by averros on Untapped Potential at 13-18 · 2022-11-22T21:39:20.821Z · LW · GW

The same argument applies to academic learning... it doesn't appeal to all.

What we need is diversity in education - scrap the system of academic accreditation and the prohibitions on "child labor".  So that more academic-inclined kids may go for academic courses while more hands-on oriented kids would join companies and learn on the job (actually there is some of that in family-owned businesses even now; unfortunately the estate taxes tend to destroy generational family business operations).

Comment by averros on The noncentral fallacy - the worst argument in the world? · 2022-11-22T21:20:13.145Z · LW · GW

The examples of objections to the non-central fallacy above rely, essentially, on utilitarian (consequentialist) argumentation.

This itself is a logical fallacy, as it presumes that some form of utility function implied in the utilitarian argument is the same as that of the person's who is supposed to be convinced by the counter-argument.

The real big problem with all forms of utilitarian arguments is that they assume both definition of utility and some arbitrary time horizon.

For example: "taxation is theft" is a moral argument (based on aversion to using offensive violence on moral grounds) which may imply some loss of "social value" (which is quite vague) due to lack of government, but may end up being much better for humanity in the longer term (governments are known to start wars).  I.e. as we start exploring further consequences of continuing existence of governments beyond obvious functions of facilitating peaceful commerce and suppression of small-scale crime (or whatever else governments are supposed to do) we may come to the point of view that existence of governments will eventually result in nuclear war exterminating the humanity.  Which is a much worse outcome than living in some "wild west" style anarchy.

Comment by averros on Untapped Potential at 13-18 · 2022-10-19T22:56:18.356Z · LW · GW

My first patent (for a hybrid analog/digital method of detecting pulse centroids for noisy signals) was awarded when I was 14... but, then, my father was a Chief Scientist in a large electronics design department and I had all kinds of electronic parts as toys since I was a toddler, and at some point he started taking me to his office so I started interacting with real engineers building real computers.  (When I've read Robert W. Wood's memoirs I discovered that he had a similar experience in his teenage years; in Surely You're Joking Richard Feynman writes about playing with real-world tech - fixing radios - when he was 12 years old).

I think the current idea of academia as a carefully isolated age-segregated bubble of theoretical learning is seriously misguided, and that guilds running apprenticeships was a much more useful form of education, at least during teenage years.  You need to develop problem-solving skills in the real world before going for theory - then you get the benefit of intuition and understanding of how the theories relate to the reality, and how to effortlessly cross the artificial boundaries between disciplines when you need to something in the real world (which has no such boundaries).

By the time I finished the university I already got one of the top civilian awards from the government for contributions to the computer industry.  Starting playing with things in the adult world as a teenager has benefits!

As for deference to authority, I never learned it.  Lack of it served me well by quickly getting me out of places where I was wasting my time (not that being fired feels good), but it all worked out OK as nobody can fire me now:)

Comment by averros on Humans aren't fitness maximizers · 2022-10-07T00:20:29.027Z · LW · GW

This completely misses the point, for a simple reason: humans (uniquely among Earth lifeforms) are subject not only to genetic/epigenetic evolution but also to memetic evolution.  In fact, these two evolutionary levels are tightly coupled, as evidenced by the very good match between phylogenetic trees of human populations and of languages.

It makes no sense to talk about human IGF, any definition excluding memetic component is meaningless.   Now, if you look at IGMF optimization a lot of human behavior starts making a lot more sense.  (It is also worth pointing out that memetic evolution is much faster, so it is probably the driving factor, way more important than genetic.  It is also structurally different, more resembling evolution in bacterial colonies - with organisms swapping genes and furiously hybridising - than Darwinian competition based on IGF.)