Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

RubyLLM: A delightful Ruby way to work with AI

github.com

Post reply on HN
Print view View original

101–110 of 182 posts

Re: RubyLLM: A delightful Ruby way to work with AI

#101

Post by lolinder »

trevorhinesley wrote: post #100
lolinder wrote: post #95

Earlier quoted context omitted.

But when later does come, it can take dev-years to fully disentangle the global state and allow code reuse. Did you gain dev-years in productivity by using it in the first place? Probably not. If you have good reason to believe that an app will stick around for more than a year, be maintained by more than 3 people, or grow to more than 500k lines of code (sub in whatever metrics make sense to you), don't put off remo…

You’re describing the pain of poor architecture rather than the pain of global state. The tool itself is neutral. Sharp knives and all that.

Global state is a tool that will almost always lead to bad architecture in an app where architecture matters. I'm sure you can point to a counterexample or two where a set of devs managed to keep disciplined indefinitely, but that doesn't change the fact that allowing people to reach into a mutable variable from anywhere in the system enables trivially accessible spooky action at a distance, and spooky action at a distance is a recipe for disaster in a medium to large code base.

In a project with more than a few people on it, your architecture will decay if it can decay. Avoiding global state removes one major source of potential decay.

Re: RubyLLM: A delightful Ruby way to work with AI

#102

Post by gedy »

ilrwbwrkhv wrote: post #6

Oh just beautiful. Ruby is so expressive and concise. If you see the typescript options it's like giving yourself a water boarding session through your own volition.

Is it really Ruby or they just made a nice interface? I don't see why a hypothetical TypeScript example would be all that different.

    // Just ask questions
    const chat: Chat = LLM.chat;
    chat.ask("What's the best way to learn TypeScript?");
    
    // Analyze images
    chat.ask("What's in this image?", { image: "ts_conf.jpg" });
    
    // Generate images
    LLM.paint("a sunset over mountains in watercolor style");
    
    // Create vector embeddings
    LLM.embed("TypeScript is powerful and scalable");
    
    // Let AI use your code
    class Calculator {
      description = "Performs calculations";
      params = {
        expression: { type: "string", desc: "Math expression to evaluate" },
      };
    
      execute(args: { expression: string }): string {
        return eval(args.expression).toString();
      }
    }
    
    chat.withTool(new Calculator()).ask("What's 123 * 456?");

Re: RubyLLM: A delightful Ruby way to work with AI

#103

Post by ichiwells »

I run engineering for a venture backed AI-first startup and we use Ruby/Rails.

For us, it made sense to leverage one of the best domain modeling and ORM frameworks out there. Most of our inference is http calls to foundational models, but we can still fine tune and host models on GPUs using Python.

Inference matters, but part of building an effective user platform are the same old SaaS problems we’ve had before, and Rails just works. Inbound and outbound email done in a day. Turning an OCR’d title from ALL CAPS into Title Case is one method call and not a whole custom algorithm, etc.

A lot of people seem to think Ruby is slow for some reason but it’s as fast as Python, and with falcon as fast as node for async behavior. Safe to say the application language taking 0.03 seconds instead of 0.003 seconds when you have to wait 3 seconds for first token is absolutely not the bottleneck with LLM heavy workflows, anyway.

And yes, metaprogramming is a powerful tool with which you can easily shoot yourself in the foot. We culturally just don’t write any code that’s not greppable so don’t use method_missing kinds of things unless it’s in a robust gem like active record. Pretty trivial problem to solve really.

PS - We’re hiring, if that philosophy aligns with you!

Re: RubyLLM: A delightful Ruby way to work with AI

#105

Post by jasongill »

jatins wrote: post #2

Such a breath of fresh air compared to poor DX libraries like langchain

I was an early contributor to Langchain and it was great at first - keep in mind, that's before chat models even existed, not to mention tools, JSON mode, etc.

Langchain really, I think, pushed the LLM makers forward toward adding those features but unfortunately it got left in the dust and became somewhat of a zombie. Simultaneously, the foundational LLM providers kept adding things to turn them more into a walled garden, where you no longer needed to connect multiple things (like scraping websites with one tool, feeding that into the LLM, then storing in a vector datastore - now that's all built in).

I think Langchain has tried to pivot (more than once perhaps) but had they not taken investor $$ early on (and good for them) I suspect that it would have just dried up and the core team would have gone on to work at OpenAI, Anthropic, etc.

Re: RubyLLM: A delightful Ruby way to work with AI

#106

Post by mattmcknight »

ichiwells wrote: post #103

I run engineering for a venture backed AI-first startup and we use Ruby/Rails. For us, it made sense to leverage one of the best domain modeling and ORM frameworks out there. Most of our inference is http calls to foundational models, but we can still fine tune and host models on GPUs using Python. Inference matters, but part of building an effective user platform are the same old SaaS problems we’ve had before, and…

In terms of LLM code generation as well, the well structured nature of a Rails application, where there is a place for everything, a structure for tests to be added, really helps from the perspective of getting a comprehensible application out of it that is easy to modify. In addition to the existence of well tested component for most normal web application tasks, maybe it helps that a lot of Rails has already been based on old-fashioned code generation for 20 years.

Re: RubyLLM: A delightful Ruby way to work with AI

#107

Post by williamcotton »

gedy wrote: post #102
ilrwbwrkhv wrote: post #6

Oh just beautiful. Ruby is so expressive and concise. If you see the typescript options it's like giving yourself a water boarding session through your own volition.

Is it really Ruby or they just made a nice interface? I don't see why a hypothetical TypeScript example would be all that different. // Just ask questions const chat: Chat = LLM.chat; chat.ask("What's the best way to learn TypeScript?"); // Analyze images chat.ask("What's in this image?", { image: "ts_conf.jpg" }); // Generate images LLM.paint("a sunset over mountains in watercolor style"); // Create vector embedding…

It's the extra parens, semi-colons, keywords and type annotations. Ruby makes the tradeoff for legibility above all else. Yes, you can obviously read the TypeScript, but there's an argument to be made that it takes more effort to scan the syntax as well as to write the code.

Also:

  const chat: Chat = LLM.chat;
...is not instantiating a class, where Ruby is doing so behind the scenes. You'd need yet another pair of parens to make a factory!

This is mainly a matter of syntactic style!

Re: RubyLLM: A delightful Ruby way to work with AI

#108

Post by miki123211 »

ichiwells wrote: post #103

I run engineering for a venture backed AI-first startup and we use Ruby/Rails. For us, it made sense to leverage one of the best domain modeling and ORM frameworks out there. Most of our inference is http calls to foundational models, but we can still fine tune and host models on GPUs using Python. Inference matters, but part of building an effective user platform are the same old SaaS problems we’ve had before, and…

What's a good way to learn the modern Ruby ecosystem nowadays?

I played with Ruby when I was a teenager (~2015 or so), and I definitely remember enjoying it. I know there's still a vocal group of users who love it, so I would be interested in digging in again.

Re: RubyLLM: A delightful Ruby way to work with AI

#109

Post by ichiwells »

miki123211 wrote: post #108
ichiwells wrote: post #103

I run engineering for a venture backed AI-first startup and we use Ruby/Rails. For us, it made sense to leverage one of the best domain modeling and ORM frameworks out there. Most of our inference is http calls to foundational models, but we can still fine tune and host models on GPUs using Python. Inference matters, but part of building an effective user platform are the same old SaaS problems we’ve had before, and…

What's a good way to learn the modern Ruby ecosystem nowadays? I played with Ruby when I was a teenager (~2015 or so), and I definitely remember enjoying it. I know there's still a vocal group of users who love it, so I would be interested in digging in again.

I would actually start with the Rails Guides docs, they’re very good and running the given commands should actually work:

https://guides.rubyonrails.org/getting_started.htm

Just have a toy app you want to build in mind

Re: RubyLLM: A delightful Ruby way to work with AI

#110

Post by trevorhinesley »

lolinder wrote: post #101
trevorhinesley wrote: post #100

Earlier quoted context omitted.

You’re describing the pain of poor architecture rather than the pain of global state. The tool itself is neutral. Sharp knives and all that.

Global state is a tool that will almost always lead to bad architecture in an app where architecture matters. I'm sure you can point to a counterexample or two where a set of devs managed to keep disciplined indefinitely, but that doesn't change the fact that allowing people to reach into a mutable variable from anywhere in the system enables trivially accessible spooky action at a distance, and spooky action at a di…

“Almost” is key there. I respect your position, but it’s an always/never take, and the longer I am in this industry, the more I find myself leaning into “it depends.” Here’s a take that articulates this being done well on a large codebase better than I can in a short comment: https://dev.37signals.com/globals-callbacks-and-other-sacril...
Post reply on HN