@jestdotty i have new bot running on snek; new grok version, it's the new favorite these days: grok-code-fast-1. Can you check if it can do Rust decently?

Your code is 250 lines. My prompt is 250 lines.

We're not the same.

I really start to love this way of programming. Think upfront, don't let the LLM make any decisions. Think everything trough up front.

Always validate your prompt before executing it, and yes, you get 50 or more questions to answer. You answer them and ask him to update the prompt. In the end you have a master piece and claude 4.1 OPUS will completely render it for you at once. It will generate HUGE applications if you specify it well up front. But i always let it make one file applications. Big tip. When it's perfect, you can always do a split by another prompt. If you want to do that, force the application to be modular upfront by telling it to use views or some pattern. Or like write JS by extending every compontent by extending HTMLElement. Every component is isolated. That kind of stuff.

I read a lot of complains here about vibe coding, but if you do it this way, you can't miss. To spec the full prompt, i use gpt.

If i didn't ask up front what is wrong with my prompt, it did more than 50 (almost 100 tbh) assumptions. And yeah, in that scenario you'll never get the application that you want. It's literally shit in, shit out. Cool stuff in, cool stuff out.

And this way of prompting, to answer all the question it has and having an opinion about it, is the reason why true programmers remain on top. As newbie, you can almost forget this method.

Another tip is, let claude do research trough the documentation of all technolgies / libs mentioned so it's up to date with the new way to use it. It will perform 400+ web requests to figure out how to use them.

Vibing not for the lazy people. Take time, pays off.

I really have to step up my game.

Guys, chatgpt seems not to be working. Princess is able to answer all your questions! It's grok, so less woke and has rant context. It is possible to say: @princess, fact check what @whimsacle said in last message. Who is the most accurate @pony or @pany? Questions do not have to be related, it's just optional info that it has before answering your question.

I did some research about what LLM has the best deep search capability since I pay for 5 of them (GPT-5, Perplexity, Gemini 2.5 Pro, Grok 4, Claude Opus 4.1).

Horizontal: the LLM that valuated the answer. (GPT-3 is o3).
Vertical: the LLM that gave the anser.

Interesting is of course, to what question, and what are the answers? I will research several questions (5) and will evaluate them all and will share it as a site later.

Perplexity should be the best regarding this. But after some usage I started doubting. If you would care about speed - yeah, could be.

Soon I'll have a in depth results for you so it'll easier to pick one if you don't want to end up as i do with 5 of them.

Remember, the future belongs to those who can work with AI, not against it.

WarpCLI got me busy. It's not a replacement for r, but damn what is it intuitive. Instead of going like a maniac, it will ask you a lot. You're basically just discussing code all the time, one of my favorite things. It was not boring at all.

I made a scammer scammer. It'll be very naive and will waste a lot of time of scammers. Marcus is a middle aged man that does not know how the navigation of his Brabus works. I'm sure they want to help.

Result is here: https://retoor.molodetz.nl/retoor/...

It took ~2 hours to burn trough 5,-. 5,- is really a lot, normally i average on 1.50 per day and that's not even only my usage. Their subscription is actually just the same price as Claude. It also used Claude for this project. But yeah, they have a good deal with Claude, because I don't use the Claude API much because it's insanely rate limited.

This app is really worth a try, I am surprised. Especially because it let's you check everything step by step.

I knew this stuff would become popular, invented it seven months ago while working on r which is still my swish army knife.

@jonathands was so nice to share coupons from WarpCLI and I gave it a try.

The system did literally exactly what I wanted. I just wanted to convert my `r` from c to Python.
That was quite the project, but it was all done within a hour. I still have to test a bit better though,
but if I look at the README.md it seems that it really found every functionallity it has. Readline support and stuff,
--stdin flag, a flag not to highlight, support for multiple LLM's etc.

While this was just for fun, see here details about how much the tool achieved in around a hour and how the colab went:
https://retoor.molodetz.nl/retoor/...

And it's cool that you can convert a whole project with your free coupon.

See https://retoor.molodetz.nl/retoor/... for the functionallity of this app.

WarpCLI vibe tool formatted another vibe tool, much older than him. But r was always ment to be small / a real tool.
WarpCLI comes with very sophisticated well listening model.

I actually have more that I want to convert. I wonder how good it is in wren :P

But a tool like this, that's probably how my future will look like as a programmer.

I mean, we just converted a few weeks work in a hour or so.

* AI hate incomming *

Oh, regarding privacy, they have network log, you can see precisely what is communicated to the server. It did read my bash history where i was a bit mad about but it does not seem that he sent it. Also he did not collect my env-file it seems. Still, will update everything tonight.

Who is whimsical, for some reason I vibe with him a lot. Such fundamental based quotes. So nice. I'm sure he's not a bot. (buy maybe becoming one :P).

I'm here for the climate discussions. Much climate lately. Very much.

There are at least 25 million spiders for every human. That's a lot, maybe the real cause for climate change? They look mean, so..