jampekka

27 ms·

451.
by jampekka 1y ago
I don't think the Chinese room thought experiment is about this, or performance of LLMs in general. Searle explicitly argues that a program can't induce "understanding" even if it mimicked human understanding perfectly b
452.
by jampekka 1y ago
This is not the same phenomenon though? Chromatography is based on different adsorption affinity of the different molecules of the liquid to the stationary phase.
453.
by jampekka 1y ago
It's just providing the most valuable commodity of friction. :)
454.
by jampekka 1y ago
Government provides the enforcement of property rights. I didn't sign for property rights, does that make them immoral? There is right to roam in many countries, meaning you can just live in a forest, although you'd have to rely o
455.
by jampekka 1y ago
A lot of philosophy is about finding, refining and arguing about definitions and that process can't be really separated from the "content". This stems largely from the very problem that it's hard/impossible to find
456.
by jampekka 1y ago
> You need to resolve what you mean by 'free'. If you have no option but to pay tax, ie your wealth is forcibly extracted, you are not free. You can't both be free and have wealth forcibly extracted. In many places you&#x2
457.
by jampekka 1y ago
I don't think this is it. There is definitely a lot of philosophy that tries (and consistently fails) to construct formal definitions of some "true ontologies". But this is more about trying to figure out reasonable definitio
458.
by jampekka 1y ago
Depends on your interests. Chalmers is probably the most influential philosopher of consciousness, and I find the most reasonable. The article definitely shouldn't be dismissed off-hand, but it may be a bit difficult to appreciate as i
459.
by jampekka 1y ago
> T he raw bits of the base pairs is just one component of the information, but it’s like a maximally compressed version of the info. This sounds a bit suspect. Maximally compressed version would be very sensitive to mutations which woul
460.
by jampekka 1y ago
Work for others owning the means of production or starve is not a fact of nature.
461.
by jampekka 1y ago
Thinking yourself as a slave if your income hits the top tax bracket is not less confusion.
462.
by jampekka 1y ago
In Athenian democracy there were more literal slaves than citizens with voting rights.
463.
by jampekka 1y ago
Policies or ideologies don't fall on some 1D left-right axis. MAGA e.g. has authoritarian and totalitarian ideas and aspirations that differ from the US government of the 1940s.
464.
by jampekka 1y ago
E.g. in feudal Norway there was no (de-jure) serfdom: https://en.m.wikipedia.org/wiki/Serfdom_in_Norway Also e.g. Finland had a manoral tenant system until the 1920's where the tenants were free to leave. https:&
465.
by jampekka 1y ago
The portrayal of medieval feudalism here is a bit of a caricature. Serfdom was not a universal feature of feudal societies and at many times and places peasants were free to go and starve on their own. This is the dynamic of the currently d
466.
by jampekka 1y ago
There are very few cultures with 0 tolerance for e.g. nicotine or especially caffeine. I'm not sure new drugs are even needed. There are plenty of existing drugs, and the vast majority are less harmful than alcohol. They would of cours
467.
by jampekka 1y ago
I find it's especially needed for socializing with drunk people.
468.
by jampekka 1y ago
Yes. None of these are a feature of the intoxicant itself.
469.
by jampekka 1y ago
Everybody I personally knew who died of alcohol drank too much alone. They also drank a lot in groups but oftentimes held the drink better than many who drank less.
470.
by jampekka 1y ago
Alcohol has exeptionally bad ergonomics when it comes to intoxicants. Its strong effects on behavior and experience are wildly unpredictable, it has severe undesirable effects, it's stongly toxic, it causes morbid dependence and has pa
471.
by jampekka 1y ago
Distro maintainers/packagers are who keep the current software stacks running. It's rather amazing how they manage to keep the billion or so lines of separately written code working in unison. That said, the labor needed to keep t
472.
by jampekka 1y ago
What needs to be "learned properly" is sadly a huge pile of incoherent legacy cruft that ideally wouldn't be there at all. This is not to denigrate the huge and critical effort that makes current computing possible, and that
473.
by jampekka 1y ago
I take bit less unstable dependencies over the total mess of C++ dependencies with CMake, shared libraries, version conflicts etc any time. There's probably also a bit of an illusion about C++ transitive dependencies due to them usuall
474.
by jampekka 1y ago
OLS is a convex optimization problem, so this doesn't really apply. And for statistical analysis you really don't want to add poorly understood artificial noise to the parameter estimates anyway.
475.
by jampekka 1y ago
Yes, and I was explicit about it in another comment in this post.
476.
by jampekka 1y ago
Not everything is a linear combination of large number of (IID) samples, and thus not everything is gaussian distributed.
477.
by jampekka 1y ago
Central limit theorem tells in practice that gaussian distributions is can be expected to be quite common. And it makes the gaussian distribution a good first guess. Least squares gives the ML estimate for gaussian residuals. I don't f
478.
by jampekka 1y ago
With batching it becomes SGD. If you're OK with approximations, you have e.g. randomized, reduced rank and streaming SVDs. And these tend have a lot nicer approximation and convergence properties than SGD. What are the common cases for
479.
by jampekka 1y ago
Computationally OLS is taking the pseudoinverse of the system matrix, which for dense systems has a complexity of O(samples * parameters^2). For some GD implementations the complexity of a single step is probably O(samples * parameters), so
480.
by jampekka 1y ago
If by stochastic methods you mean something like MCMC, they are increasing in popularity, but still used a lot less than analytical or numerical methods. And almost exclusively only for more complicated models than basic linear regression.

More ›