As someone who does research with computer algebra systems (mainly Wolfram Mathematica) many hours a day, I think that SymPy is still far behind Mathematica. The fundamental design of Mathematica based on replacement rules and a style of functional programming is quite difficult to beat in terms of ease of use. A few weeks ago I gave SymPy another try, and tried to implement a nested sum of the type 1 <= a_1 < ... < a_k <= n over a complicated summand f(a_1, ... ,a_k), where k and n are integer arguments, and it took me ages to figure out how to implement this in SymPy, whereas I in Mathematica I can write this in a minute.
That's true, Mathematica is more advanced. But it's not cheap and it brings in a whole new world with it. You do have to learn to use it explicitly while with SymPy, you can start after a 5 minute introduction. I would say, Mathematica a good professional instrument while SymPy is a good hobbyist tool.
A note to a bypassing reader, if you want to try Mathematica, and not sure if you would stay with it, don't go for a desktop license. Get a Raspberry Pi instead. Mathematica comes free for Raspberry!
SymPy is a great tool but there isn't a dedicated team of mathematicians and programmers working on it full time. I believe it has potential to compete with Mathematica someday.
I'm a heavy user of Mma but there are better examples. Automating the rewriting of formulas to make them more concise is a big one. (But also not as straightforward as it could be, tradeoffs between conciseness, ease of reading and speed of implementation need to be triaged, similar to Lisp/APL. Maybe sympy still has a chance to be more like math/TCS)
No. In Mathematica important operations like mapping functions over sets (arrays/tensors) at various depths have simple operators such as (/@, @, @@, @@@). You can do functional composition using @*, etc. You don't need to define your variables and functions before being able to write them in expressions. Often you want to have intermediate terms that have some temporary name, and which get substituted later on. The replacement rules in Mathematica are very powerful for this.
As another example, I googled some project Euler solutions in Mathematica and (pure Python) and found this: https://www.nayuki.io/page/project-euler-solutions. Compared to the pure Python solutions, the Mathematica code is typically much smaller. Of course this is not Sympy, but a lot of the Python syntax carries over to SymPy as well.
I don't have experience with SageMath, but if they are really running on Python, then they are surely giving up on some of the nice rewriting that Mathematica is based on, as I would expect then a library call is needed instead in Python.
> As someone who does research […] mainly Wolfram Mathematica […] it took me ages to figure out how to implement this in SymPy, whereas I in Mathematica I can write this in a minute.
Hardly surprising, is it? As someone with more experience in Python, it would likely be the opposite for me.
You are replying to somebody talking about sage (rather than sympy), which in many fields (not all) is very far ahead of Mathematica.
I do quite like the programming language of Mathematica, but an advantage of sage (and of sympy) is that you have the whole python standard library at your disposal. For your example you could e.g. use itertools.product.
From my personal experience I have seen this disparity in the extreme, since I studied physics where the gender ratio was very unbalanced. I knew many guys who had some trouble socializing, but were very nice people, and they did not get any attention from women.
He is saying that the "base code" / DNA which makes up the brain might be only 40MB (for example). Not that what eventually emerges from that (such as our memories and learned abilities) can be captured with just 40MB. It's similar to how the Game of Life can be implemented in just a few lines, but very complex behavior can emerge from that. The key is to find a sufficiently simple but general model from which intelligence equal to our own can emerge given sufficient training.
I understand that, but that is an extremely banal observation if you think about it, because the fact that there is this incredible emergent behavior from a simple starting system is the heart of the mystery here.
One of the things that everyone is sort of skipping over is the "sufficient training" part. There is no bootstrap reinforcement learning possible for AGI. You can't alphago this sucker and have it play simulations against itself because the whole idea of generality is that there isn't a simple rule framework in which you could run such a simulation. So training any kind of AGI is a really hard problem.
hes specifically answering the question of why he thinks he has any chance of success doing this independently when there are giant organizations funding this.
He admits that the equivalent of years of "training" would still be needed to take an toddler-level consciousness to something approaching an adult human.
> GPT-3.5 series is a series of models that was trained on a blend of text and code from before Q4 2021. The following models are in the GPT-3.5 series:
code-davinci-002 is a base model, so good for pure code-completion tasks
text-davinci-002 is an InstructGPT model based on code-davinci-002
text-davinci-003 is an improvement on text-davinci-002
You're totally true. But it turned out, that proper prompting (like packing dialog context into the prompt) worked great.
I spent many hours with orig ChatGPT and with this recreated version. The main difference I have found is that the recreated ChatGPT is more inclined to ask questions on questions (maybe can be fixed with more prompt engineering). I didn't find any major differences in the quality or usefulness of the answers.
Even that might not be entirely safe, as Schrödinger's behaviour has only become widely known in recent years for example. (Although he died around 60 years ago, so not quite 100.)
I have not used Sunvox but I make music as a hobby using trackers like Renoise, and also various DAWs. I would still say that there is an advantage to having a specially designed hardware synth with intuitive knobs and faders. I'm glad that KORG can make reasonably affordable and professional synths while using off the shelf hardware like the Pi. In my opinion the presets and ease of use of paid synths (both software and hardware) are still quite a bit ahead of free options.
I don't want to make SunVox or its other fans look bad so I just encourage anyone reading this to give it a whirl.
It's jam packed with features and stellar patches/examples despite its simple and clean interface. It really is a joy to work with, especially on a tablet sized screen. It's so good when scaled properly that I prefer it over physical controls sometimes.
You may or may not like the pattern editor because it's a tracker interface, but the rest of the tool is so good live that you can initially ignore it and then slowly incorporate it into your workflow.
Not the person you were replying to, but that takes me back! Back in those days I was in early high school and Jumper was one of the games I looked up to when I was learning Game Maker. I completely forgot about it.
