I doubt if it is even possible for anyone to make rigorous claims about broad software quality over long periods of time. The common discourse over this subject merely consists of sharing anecdotes and speculative pet theories. To answer this question convincingly one would ideally need to take a comprehensive and representative sample across the whole range of software from the past and today and make comparisons across the vast categories of software that exist. Doing this even for today’s software would be a tremendous undertaking for a variety of reasons. Now consider doing it for the past where one must face the simple fact that the vast majority of historic software is lost or incomplete. Even if we have binaries we might not have the source code. Next we may not even have a computer to run the software on. The software that does survive might not be a representative sample of past software-- it may have experienced survivorship bias due to its quality or for some other reason.
The vast majority of software (past or present) is not distributed and therefore difficult to access. So I find when people talk about “software” in this context they often mean commercial “boxed product” software or software that otherwise has an end-user that is the general public rather than “all software”. That is simply the only neighbourhood of the software universe that most people are familiar with. Yet even if we were to restrict ourselves to that category of software there is a complete lack of convincing comprehensive comparisons. And even if there were it might only be an indication of a trend confined to that category of software.
Most commentators suggest the trend concerning software quality continues and may even accelerate in the future. So instead of trying to compare the present with an inaccessible past, why not compare the present with an eventual future? Design objective metrics and/or recruit esteemed judges for subjective assessments, then assess today's software and repeat this assessment a number of years later and compare the metrics.
> This release introduces the SFNNv16 network architecture, reducing binary size by removing redundant threat features…
This is nice to see. One of my few gripes with NNUE engines is their huge size compared with the old “HCE” engines I was accustomed to (which were usually hundreds of KiB or low single digit MiBs in size). I will trade off size for superior playing strength but at a certain point it becomes hard to keep justifying a 2x increase in size for what seems like ever diminishing practical returns. I am still searching for a strong NNUE engine that is ~10MiB or less. I have found Marvin, which is small at 1.8 MiB, but it is nowhere near as strong as Stockfish.
I have been wary of the risks surrounding AI since watching Robert Miles on Computerphile more than a decade ago. Since then I have paid at least some attention to developments coming out of AI safety research. Even back then the discussion was fraught with doubt which, given the nature of this topic, was to be expected. It is unfortunate that it is worse now, when the stakes are much higher.
There is no fundamental difference between symbols inscribed on a stone tablet with a sharp tool versus symbols etched onto a reflective disc by a laser. In both cases you are recording symbolic information on a physical medium; they both record digital information.
The true difference between digital and “analogue” lies in the nature of the representation and interpretation of information, that is whether they are descriptions (symbolic) or analogies. The shape of the groove on a phonograph record is an analogy of a sound wave signal. The pits in a compact disc represent numerals which give a description of the amplitude of a signal at a given point in time.
> Maybe if we had some versioned web standard that lagged behind the "real" ones so sites could say "StableWeb 2.0 Compatible!" or something.
There have been a few attempts at this already and I have something like this myself for websites I design. Ultimately they fail to gain widespread traction since everyone has very different ideas on what of the various Web standards should be included at each version level. Someone else’s standard will likely not include that one feature you cannot go without so you simply do not follow it.
Thankfully for me it happened a while ago, after chess was conquered. One of the most amusing (and depressing) things in the past few years of the current “AI era” is people making the same claims and having the same discussions about “intelligence” and “creativity” from ten, twenty or thirty years ago as their field is finally coming under attack by AI; they are finally going through what the competitive board game communities (checkers, Othello, chess, shogi, go etc.) have already come to terms with. There was no single moment but a gradual process over decades as the incursion became greater and greater, forcing a recalibration as previously held positions became indefensible. In my own case it spurred me to educate myself in many areas of mathematics, philosophy and neuroscience in particular where I was ignorant and has undoubtedly shaped my understanding of “intelligence”.
When people started claiming LLMs could not produce ideas outside of what they encountered in their training data, I was reminded of an old Chess Life article from the 60s reporting on the first chess computer to play in a tournament at a local chess club. One of the club members remarked that he did not believe that the computer could play moves that were not “put in” the computer in advance. There is an overwhelming sense of “here we go again”.
https://wikipedia.gptzero.me/
reply