You are viewing archived messages.
Go here to search the history.

Arvind Thyagarajan 2025-04-09 16:07:48

Has anyone played extensively with the swathe of "speak english into code" systems out there, from idea through to deployment to an appreciative public? (I haven't...)

I'm wondering if, at the end of the day, you still have to be somewhat of a textual code / software engineering native in order to take advantage of this particular present-future of coding? Or if it's truly democratising access to practical computation (i.e. domain specific to the domain of semi-commercial tools, apps, integrations that software engineers work on)

I feel a lot of public excitement around it that I'm not feeling -- but as someone keen on encouraging everyone to act computationally for themselves, public excitement is * the metric so I'll admit to being mildly envious 🙂

Jason Morris 2025-04-09 16:17:08

I feel like it works for a small subset of what a native code speaker can accomplish with the same tools. But for people who are not native code speakers, that's a very large increase over what they could do before. I've seen people use it successfully to create a one-off web UI for a data annotation task, rather than type values into a column in a spreadsheet, and be DELIGHTED by that experience. So it is maybe both very limited and very empowering.

Mariano Guerra 2025-04-09 16:19:58

I would like to see a "pro vibe coder" in action to see what it looks like when done right, if anyone has a youtube link or similar please share.

Scott 2025-04-09 16:47:17

I've had a lot of success using bolt.new and combined with some techniques from this vibe coding doc: github.com/swerner/vibe-coding like taking a short prompt, having Grok or one of the thinking models make you a design doc and just throwing that right into Bolt for a first pass

I also released this kind of silly Ruby Gem this weekend called Monkey's Paw (worksonmymachine.substack.com/p/introducing-monkeyspaw-a-prompt-driven) where you write descriptions of what you want in markdown files and that's used to render pages for you...

📝 bolt.new

Prompt, run, edit & deploy apps

📝 Introducing MonkeysPaw - a prompt-driven web framework in Ruby

Be careful what you wish for...

Scott 2025-04-09 16:48:28

x.com/ProgramWithAi does a lot of live streaming on the weekends and might be a good example of the "pro vibe coder" you're looking for

🐦 Manuel Odendahl (@ProgramWithAi) on X: Real-world programming with AI. No filler, just killer! Unique advice and creative techniques that you won't find anywhere else, straight from the trenches.

Scott 2025-04-09 16:50:37

Also able to chat more about any particular topic if any of you are curious? Been pretty deep in all of it from the starting point of a long time "software engineering native"

Arvind Thyagarajan 2025-04-09 20:56:29

from the starting point of a long time "software engineering native"

this is why I'm curious what the experience is actually like for those who aren't

does it feel uncomfortably mysterious but just in a different way? does it feel invitingly capable, then you get hit with some uncanny valleys, and then you're lost again except with some code in your hands that's illegible to you? does it feel unsafe, like you've caught a swamp alligator, and now... what about vibe leaks and vibe attacks and vibe overflows 🤷

i want to be sure to say this isn't snark or criticism -- i'm genuinely wondering how intellectually open and inviting to outsiders this new scene really is when it comes to getting "real work" (open to definition) done.

Paul Tarvydas 2025-04-10 02:57:26

The idea that operations are written vs. spoken is orthogonal to the act of programming. I believe that the 1% will not find much use for LLM-based programming, regardless of input method. I do, though, think that the 99% will find LLM-based programming to be a useful approach for creating "good enough" software for their needs.

... programmingsimplicity.substack.com/p/llm-based-programming?r=1egdky

📝 LLM-Based Programming

2025-04-09

Jason Morris 2025-04-10 03:22:35

bsky.app/profile/hillelwayne.com/post/3lmgjrjzis22w Hillel is in the 1% and evidently disagrees...

https://bsky.app/profile/hillelwayne.com|@hillelwayne.com: Vibe coding makes it really clear how many amazingly useful programs fall in the 50-100 line range and the only reason I hadn't made them already is that I didn't want to learn the libraries required for 40 of those lines

Paul Tarvydas 2025-04-10 19:28:17

I wonder if that thread contains evidence to counter both extreme viewpoints? It certainly overlooks a much bigger problem, though. Firstly, the example of the junior programmer struggling with vibe coding. The junior was not yet a 1%er and was not benefitting from the use of vibe coding - this counters the extreme viewpoint I was espousing of 99%ers benefitting from LLM-code-generation. The senior programmer - 1%er - showed the junior how to completely skip using vibe and slapped together a 1-liner power shell script. That counters the other extreme viewpoint - there was a simpler way to approach the problem, but the LLM wasn't used. The much bigger problem is: software development is too complicated (alarmingdevelopment.org/?p=766). Even the act of writing small programs is discouraged by what we've got. IMO, smart people should be fixing that problem instead of simply adding to it. LLMs are trained to keep creating complicated software, i.e. LLMs are not much help on that front. [I have strong opinions about the issue of simplifying software development. IMO, bloatware is being caused by forcing ourselves to use one and only one paradigm for all software, i.e. the over-use of blocking "functions", instead of, say, something with a lot less coupling, like relations or SCNs (nano-DSLs) or Parts Based Programming (0D) or ???].

📝 Getting to simple

There is one gigantic problem with programming today, a problem so large that it dwarfs all others. Yet it is a problem that almost no one is willing to admit, much less talk about. It is easy to i…

Jack Rusher 2025-04-10 16:48:33

For anyone working with LLMs (from my colleagues at the lab where I’ve been helping out the last few months):

arxiv.org/abs/2504.07081

📝 Self-Steering Language Models

While test-time reasoning enables language models to tackle complex tasks, searching or planning in natural language can be slow, costly, and error-prone. But even when LMs struggle to emulate the precise reasoning steps needed to solve a problem, they often excel at describing its abstract structure--both how to verify solutions and how to search for them. This paper introduces DisCIPL, a method for "self-steering" LMs where a Planner model generates a task-specific inference program that is executed by a population of Follower models. Our approach equips LMs with the ability to write recursive search procedures that guide LM inference, enabling new forms of verifiable and efficient reasoning. When instantiated with a small Follower (e.g., Llama-3.2-1B), DisCIPL matches (and sometimes outperforms) much larger models, including GPT-4o and o1, on challenging constrained generation tasks. In decoupling planning from execution, our work opens up a design space of highly-parallelized Monte Carlo inference strategies that outperform standard best-of-N sampling, require no finetuning, and can be implemented automatically by existing LMs.

Konrad Hinsen 2025-04-11 06:38:05

Planner/Follower... Now I am waiting for the first postmodernist manifesto calling for deconstructing the power structures of LLMs. Probably written by PhiloGPT.

Ezhik 2025-04-13 13:35:15

I wanted to learn how the CodeMirror editor works, so I asked GitHub Copilot's agent mode to build a little app that uses it for me. We had a back-and-forth where I pointed out the bugs I've seen in the UI and then it fixed it. But all this took place without me having to look at the actual code, so I don't really feel like I learned anything from this experience. That makes me wonder - do people here use these tools for learning, and if so, how?

Mariano Guerra 2025-04-13 13:36:02

by looking at the working code and taking what I need 🙂

Ezhik 2025-04-13 13:37:07

That's how these AI tools learn too!!

Ezhik 2025-04-13 13:38:16

It's kind of the best I've got so far - either asking it for code samples or skipping the middle man and just looking through the source code for various projects to see what makes them tick. But I do wonder if there are better ways to do this.

Ivan Reese 2025-04-13 17:56:14

I haven't found agents to be a net positive use of time or energy in the handful of times I've tried them.