Wolfram Notebook Assistant | Hacker News Reader

Wolfram Notebook Assistant

writings.stephenwolfram.com

97 pointsnsoonhui37 comments

I loved mathematica during university. Gorgeous math notation written with quick hotkeys, so I never needed Latex. And it computed in that form, so I never needed to code in ASCII. I'd do most of my math/engineering/physics homeworks in it. I would have paid for this back then for sure. Sadly, I rarely need to open mathematica anymore, in the world of real life employment.

Same here, truly the most intuitive analytics program. The notation made me want to use it every day. I loved the animations/charts you can produce.

Why is that? Does it lack features for real life usage? Or is it too expensive?

It's probably the fact that it costs money at all. Python libraries and Jupyter notebooks can technically do a lot of things for free, so even rich tech companies just default to the free option. Getting approval for paid software is a huge pain in the butt. Also, the fact that it doesn't use simple ASCII character code is probably frowned upon since you can't diff it in a version management system for code review.

I do have a personal copy and I definitely do open it up when I really want to solve a hard mathematical problem quickly. Then I translate it after the fact to whatever corporate language I was supposed to solve it with.

That looks nice. For the user, however: If notebooks or anything connect to Wolfram servers, is your own IP mined to power the training? Has past IP been mined?

"Useful to the Point of Being Revolutionary" is an insane title for a blog post like this.

Agreed. However, it's also the baseline expectation for anything from Wolfram, though. Hyperbole is the brand.

Why? It seems really useful, so there must be a bar where that becomes revolutionary.

But it's also a valid descriptor for Mathematica :-)

... for a significant slice of professional and hobbyist use cases, and there are no alternatives AFAICT!

I really appreciate the depth and detail of Wolfram's blog posts like these. Any other blog would have ended around 1/5th of the way. I kept expecting it to end, but it kept on giving and throwing in more examples. It's amazing and I can't help but compare that most people don't do that.

1/5 of the way he was still repeating how revolutionary this is. it's also where I closed the tab.

I like the idea of this quite a lot (at a minimum, it would cut out all of those "how do I do XYZ in Mathematica" web searches) but not enough to subscribe. I'd rather buy some bucket of credits because my use of MMA is very unevenly distributed over time.

My feeling as a mathematician is that adding gen AI to study science takes the beauty away from scientific discovery. It is an overt signal to what was alreadly happening covertly: the transformation of science into a mechanized procedure for the sake of production instead of human good. I condemn technology like this especially.

I disagree. AI used thoughtfully as a tool can potentially allow a mathematician or scientist to progress further and see more beauty. I'm also a mathematician by training and I've realized that a large part of my love for math was/is the process of solving a problem by my own wits. AI does take that away somewhat, but that's not really the important part of maths for society as a whole.

Awesome work

The use of 4 times "leverage" (and forms of it) in a single article is somewhat disappointing (discounts real technical content)…

Wow… incredibly expensive… at minimum another 26€ a month. Wolfram really enjoy milking their customers dry (and I’m a customer). I think this is highway robbery.

With a bit of creativity you can run something similar for free. It's worth noting Raspberry Pi comes with a free Wolfram license. The assistant part can come from something like Perplexity, which in my experience is decent at writing Mathematica code.

I also have a Mathematica license which I think is worth paying for. One of the few closed software that has no good equivalent in the libre world. Nonetheless, I wish Wolfram had come up with a different business model that made Mathematica more mainstream, as I feel it has not realized its potential in certain areas.

Totally agree that they have unrealized potential. It was amazing when I was in grad school. Then when I started running a corporate data science team, I struggled to find a use case to recommend it for purchase. There are so many other tools that are better at loading a messy dataset, cleaning it, doing EDA, building a model, and productionizing and monitoring the model, which is the process we spend a lot of time doing.

It's still the very best for symbolic math. But it has fallen behind on other domains.

And even in the symbolic math niche, I am afraid that Python and Julia will slowly erode its market-share.

I wouldn't call that "incredibly expensive". Certainly for what it does, there are many other software packages that cost more.

For somebody who uses Wolfram Home, it’s incredibly expensive.

Because of what logic actually?

Are they forcing you to use it?

Or do you expect them to run expensive GOUs for you for free?

Are you actually using wa? Because if it's helpful as an assistant and it only costs 26 Dollar, shouldn't you be happy about this?

Yes I do use Mathematica. 26€ a month is pricing like a freakin’ luxury.

It’s only expensive if you’re getting less than 50$/month of value from it, at which point don’t subscribe. This isn’t something I’d pay for right now, but when I was doing this kind of work regularly it would have been an easy purchase.

Subscription models don’t mesh well with casual users. However, if it’s going to save you hours of work every month then it’s a perfectly reasonable price point.

Would you sub for $50 if you were getting $51 per month of value?:)

Any alternative where you pay $10 for $15 value or you pay $100 for $150 value will beat this

> Would you sub for $50 if you whee getting $51

Personally no, 50$/month of value covers not just 30$/month, but also the overhead of yet another subscription.

So paying 10 for 15 doesn’t make much sense IMO, but $100 for $150 does.

Yep. It is not black and white

That’s like two Starbucks visits.

(Wolfram customer since 1988 too, so I think it’s helpful to change units, in this cost evaluation!)

For me, it's like days, weeks, or months of distraction trying to get the license renewed, through the multiple bureaucracies that are involved: IT, purchasing, and accounting, to name a few. And a second license for the lab, or a site license, or one for home use? That's crazy talk. And now that software is sold by subscription, it's an annual headache, per app.

I get it that "single user at single computer" is the majority of use cases, so I'm not asking for accommodation. But for me, free means free site license, and I do like to have my tools installed on every computer that I touch.

Sure, there are work-arounds, but it's hard to keep using the proprietary stuff when the free stuff is so easy to deal with.

Disclosure: Wolfram user from 1993 to 1997. Also, when colleagues want to use proprietary software tools, I always go to bat for them, to get the expense approved, because I believe that people should have a choice.

Yeah i totally do understand what you're saying. I miss having mathematica available on any machine for instance on a campus site license, which i used for years overlapping yours too!

There's another way to do this:

English -- (LLM) --> python

Human fine tunes python

Python -- (py2many) --> smt2 --> z3 --> result

Perhaps some steps can be eliminated in the future.

With an example: https://gist.github.com/adsharma/45fbb065a8fe793030e8360daeb...

You could have a puzzle that's described in plain English. I think the current day tech is sufficient to translate the puzzle to a python constraint like in this gist.

The argument is that the SMT solvers are a better match for solving these problems than expensive matrix multiplication paid for by someone else.

Of course, SMT is not a solution for every situation that Wolfram language is used for today. But a lot of the reasoning benchmarks fall in this category.

There is a wolfram blog post comparing python vs wolfram from 2021:

https://blog.wolfram.com/2021/11/04/lessons-learned-migratin...

The linked tweet (https://x.com/WolframResearch/status/1325935891651698690?s=2...) is inaccessible. Does anyone have an archive?

I'd love to see pandas/df compared instead of the iterative style.

> There is a wolfram blog post comparing python vs wolfram from 2021

Now the view from someone who quit Mathematica walled garden to embrace the Python ecosystem (2018): https://paulromer.net/jupyter-mathematica-and-the-future-of-...

The same thought process as installing GNU/Linux on a random PC bought down the mall, or getting home with macOS/Windows already pre-installed.

This sounds like an argument for a custom DSL designed for the purpose of expressing constraints and for querying statiatical data.

Perhaps there are cases where DSL is the right answer. Not sure if this is one though.

I see some parallels to the arguments around using a ML derived language instead of adding types to python3 and giving it 90% of the capabilities.

There are other factors like: open vs closed development model, number of implementations (Wolfram language has exactly 1?) etc.

From my perspective, the problem to solve is that the python ecosystem isn't even trying. Would love to see stats about the adoption of the match statement introduced 3 yrs ago.

Reply on news.ycombinator.com