Darren Shepherd

39

Marc-André Moreau@awakecoding·8h

@ibuildthecloud Is it still crazy hard to capture screenshots from Wayland? I remember how zero apps could capture the screen in Wayland initially, because you had to write your own compositor or something, and it was by design

English

67

Darren Shepherd@ibuildthecloud·9h

That's an understatement omar.yt/posts/wayland-…

English

1

2

1K

Darren Shepherd@ibuildthecloud·5h

@disheng not X or wayland?

English

0

23

disheng su@disheng·6h

@ibuildthecloud My own desktop environment written from scratch already in good shape now, I can run chrome

English

0

28

Darren Shepherd@ibuildthecloud·6h

i don't even want to run kimi tbh. I just want codex or claude unlimited. So buying hardware doesn't help. I'm totally of the belief that local models don't make sense if you are talking about large ones, because by the time you have hardware fast enough it's cheaper to buy tokens.

English

0

2

924

Marc-André Moreau@awakecoding·7h

@ibuildthecloud Why not look for various inference providers for Kimi K2.5 on openrouter.ai? You'll realize you need to burn a looooot of tokens before buying the hardware to self-host the same model in its full form becomes worth it financially

English

0

5

1.1K

Darren Shepherd@ibuildthecloud·7h

I'm so sick of dealing with token usage. Can I just buy an NVIDIA rack and put it in my garage? I am curious what it would cost to run something like kimi full strength. Obviously a lot, but still curious.

English

16

1

27

7.2K

Darren Shepherd@ibuildthecloud·9h

@0xluffy You merge? I just push to main.

English

2

27

luffy@0xluffy·1d

when is the last time you actually read code you merged

English

23

0

43

9.3K

Darren Shepherd@ibuildthecloud·9h

@CodexHere @michiokage No there's nothing wrong with the process or the idea, it's that they said the phrase to begin with.

English

1

4

ℂ𝕠𝕕𝕖𝕩, ℍ𝕖𝕣𝕖@CodexHere·15h

@michiokage STUDYING is not low iq. "Reinventing" is.

English

0

1

21

ℂ𝕠𝕕𝕖𝕩, ℍ𝕖𝕣𝕖@CodexHere·15h

"from first principals" is a low IQ indicator btw.

English

0

3

76

Darren Shepherd@ibuildthecloud·9h

@jessfraz Or vibe coded 😅

English

0

1

119

Jessie Frazelle@jessfraz·1d

I don’t give a lot of compliments but the way Codex cli & app is designed so it reuses code (that anyone else can reuse as well) is really quite elegant. Elegant in a way that it could only be the result of an experienced & thoughtful team. Which is rare at their speed & scale.

English

15

10

220

22.7K

Darren Shepherd@ibuildthecloud·9h

Claude is trying to keep up with openclaw now I guess. code.claude.com/docs/en/channe…

English

281

Darren Shepherd@ibuildthecloud·9h

@leetllm Sure, sure

Français

25

LeetLLM.com@leetllm·9h

@ibuildthecloud been writing code manually for 15 years. vibecoding only breaks down if your architecture is garbage, keep your agents isolated with git worktrees and it scales way better.

English

0

1

29

Darren Shepherd@ibuildthecloud·20h

There's so many things wrong with vibe coding. Learn the skill, there's advantages. But there are also huge disadvantages. I'm telling you, the future of coding is going to look very close to what we do now, just with some better tools and approaches. It's a step improvement that will have a compounding effect. Not a hard switch to an unrecognizable future.

English

9

0

25

2.2K

Darren Shepherd@ibuildthecloud·10h

@thdxr Yep, but I don't think the black boxes will survive because of people like you.

English

3

391

dax@thdxr·13h

i wish there was more information about this like what model they started with feels weird that all of a sudden the tools we're relying on are super opaque hasn't been like this for decades

Cursor@cursor_ai

Composer 2 is now available in Cursor.

English

50

2

593

61.7K

Darren Shepherd@ibuildthecloud·10h

but actually... it does it well. I know this only because I've used chatgpt to program roblox games which is all lua. The language is so simple it's kind of hard for it to get it wrong. But a big problem is that you want tools the validate the code and the no types makes that hard.

English

34

Ellie Huxtable@ellie_huxtable·10h

@ibuildthecloud lolol too true

English

0

1

34

Ellie Huxtable@ellie_huxtable·1d

i wonder how well agents write lua because lua sandboxes super well and has for a long time see luau

dax@thdxr

we've been experimenting with getting rid of the bash tool agents can write js fine which can do what bash can (though some gaps with things like git) and is more cross platform and then could run that in this

English

9

0

51

7.4K

Darren Shepherd@ibuildthecloud·10h

@mjackson Codex is fine German engineering.

English

207

MJ@mjackson·12h

Use Codex

dax@thdxr

opencode 1.3.0 will no longer autoload the claude max plugin we did our best to convince anthropic to support developer choice but they sent lawyers it's your right to access services however you wish but it is also their right to block whoever they want we can't maintain an official plugin so it's been removed from github and marked deprecated on npm appreciate our partners at openai, github and gitlab who are going the other direction and supporting developer freedom

English

12

5

132

22.8K

Darren Shepherd@ibuildthecloud·10h

I think windows is actually a good platform for a openclaw style agent. I'll just leave that statement completely unsubstantiated.

English

1

441

Darren Shepherd@ibuildthecloud·10h

support openai, github and gitlab as they let you use your subscription how you choose.

dax@thdxr

opencode 1.3.0 will no longer autoload the claude max plugin we did our best to convince anthropic to support developer choice but they sent lawyers it's your right to access services however you wish but it is also their right to block whoever they want we can't maintain an official plugin so it's been removed from github and marked deprecated on npm appreciate our partners at openai, github and gitlab who are going the other direction and supporting developer freedom

English

10

1.3K

Darren Shepherd@ibuildthecloud·10h

@ArcanineXBT yep, i do. this codebase discobot.ai is 100% written by AI. nothing by hand. Check it out.

English

0

1

26

ArcanineXBT@ArcanineXBT·11h

@ibuildthecloud @ibuildthecloud do you even vibe code brah? i'm actually building Claude next generation

English

0

30

Darren Shepherd@ibuildthecloud·12h

AI-generated data. Data at this point is not exactly a problem. The problem is now about curating the right type of data. There's basically no end in sight right now for LLM. There's a difference between intelligence and usefulness. People want to argue that LLM is not intelligent or it's not smart or it's not getting smarter. Those could be true. It's very subjective. What is far more important is the utility of the LLM: is it useful? Does it provide a function that provides value? Everything points to yes but I will say we haven't quite nailed it. We're very quickly addressing all of the key issues.

English

2

Scept Meerkat@Sc_Meerkat·13h

@BrownCoyoteStu @ibuildthecloud What exactly is not true? StackOverflow was used as training data. It is now dying. People do upload LLM-generated code to GitHub, another source of training data. What do you think the next generation of LLM is going to be based on?

English

0

23

Darren Shepherd@ibuildthecloud·14h

@zeeg I really hate things like this. "Oh look, they aren't actually intelligent." Yeah, we know, they just mimic things. But that's super useful.

English

95

David Cramer@zeeg·16h

1) not surprising whatsoever 2) this is exactly what I keep saying about models not being powerful enough today the fact that they can do so much with lossy compression is amazing, but there's no magic here imo (for transformers) context windows need to be 1-2 orders of magnitude larger for the future people keep saying is reality, and even then the compute is probably not worth it

Lossfunk@lossfunk

🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵

English

14

4

94

8.3K

Darren Shepherd@ibuildthecloud·14h

ugh, i created such a mess.

English