Arpad Toth

573 posts

Arpad Toth

Arpad Toth

@arpadietoth

Katılım Kasım 2022
65 Takip Edilen67 Takipçiler
Arpad Toth
Arpad Toth@arpadietoth·
@Zhuinden @hasen_95dx C'mon.. you don't need "clean" architecture for that. I can write platform-independent code in C without any classes, weird abstractions and ceremonies. If that's the reason, it's the most inefficient way to write platform-independent code 🤣.
English
1
0
0
70
Gabor Varadi
Gabor Varadi@Zhuinden·
@arpadietoth @hasen_95dx Theoretically, the idea would have been that the software "exists in isolation to any frameworks" and could be used as a "core" for the app to run on any platforms, and all platform-specific functionality would be integrated in through adapters. KMP makes this promise too, but?
English
1
0
3
131
ハセン حسن
ハセン حسن@hasen_95dx·
"Clean Architecture" It's all glue code that you could have just automated, but instead you have to manually type it out for every single code path.
English
7
1
19
2.9K
Arpad Toth
Arpad Toth@arpadietoth·
@KaiXCreator all the things you listed, since Claude Coude cannot do any of those.
English
0
0
1
21
Kaito
Kaito@KaiXCreator·
If Claude Code can now write code, run the app, debug errors, review pull requests, and even fix bugs automatically what exactly is left for junior developers to do?
English
81
1
70
9.3K
Arpad Toth
Arpad Toth@arpadietoth·
@KaiXCreator LLM or AI? LLM are not inteligent at all.. so yes.. the limit is zero. Is as inteligent as your sofa
English
0
0
0
8
Kaito
Kaito@KaiXCreator·
Is there a limit to how intelligent AI can become?
English
121
3
79
6.5K
Arpad Toth
Arpad Toth@arpadietoth·
@Zhuinden @hasen_95dx I haven't seen a good implementation of Clean Architecture because "Clean Architecture" it's as clean as a public toilet at the end of a busy day in a train station. Just a bunch of shitty classes with no reason to exist beyond the butterflies in the author's head
English
1
0
0
159
Gabor Varadi
Gabor Varadi@Zhuinden·
@hasen_95dx I haven't seen a correct implementation of clean architecture, that wasn't just "data-domain-presentation layering" marketed as "clean architecture", in a very long time. Which is genuinely surprising in the sense that it's well-known that layer-based modularization isn't good.
English
2
1
12
1K
Arpad Toth
Arpad Toth@arpadietoth·
@Zhuinden @hasen_95dx "Clean architecture" got popular because of the name "clean" and the need of many developers to feel superior not due to their achievements but due to crappy elitism. The stupider a dev is the more he likes things like "clean" or whatwever "architecture".
English
0
0
2
66
Gabor Varadi
Gabor Varadi@Zhuinden·
@hasen_95dx "Clean architecture" got popular because it was much easier to justify additional billable work hours under the guise of "higher quality", for even the simplest feature requests Clean Arch motto is simple: "why write the code in 10 lines, if you can do it in 10 classes instead?"
English
4
3
125
9.3K
ハセン حسن
ハセン حسن@hasen_95dx·
"Clean Architecture" Every single change has to touch 10 layers & possibly rename multiple files.
English
48
54
1.7K
118.2K
Arpad Toth
Arpad Toth@arpadietoth·
@unclebobmartin I didn't saw a single company or proffesional dev who is not writing code by hand till this day and I work with lots of companies and lots of devs. Were did you get that from?
English
0
0
3
120
Uncle Bob Martin
Uncle Bob Martin@unclebobmartin·
Authoring code by hand HAS GONE AWAY. Engineering module structure and architecture has not.
English
134
136
1.4K
87.9K
Arpad Toth
Arpad Toth@arpadietoth·
@plainionist quick reminder, you can't really express all context in English, that's why we invented programming languages that are context free languages. To express all context your "prompts" will look like legal documents, very hard to write, very hard to read still interpretable
English
1
0
2
54
Seb
Seb@plainionist·
Quick reminder: LLMs cannot read between the lines. You have to provide all relevant context explicitly 😉
English
25
1
35
1.9K
Hossain Kabir
Hossain Kabir@awarehossain·
True, AI isn’t at the point of replacing developers end-to-end. It can generate code, speed up workflows, and reduce boilerplate. But it still struggles with system design, real-world constraints, and long-term maintenance decisions. Right now it’s more of a force multiplier than a replacement.
English
1
0
0
267
SumitM
SumitM@SumitM_X·
As of now , AI isnt good enough to replace developers...
English
31
10
120
10.1K
Vlad Mihalcea
Vlad Mihalcea@vlad_mihalcea·
At this point, AI has become so useful for tackling tedious tasks that there is no turning back to the days before AI. If AI vendors increase prices, many will migrate to local LLMs or cheaper alternatives rather than abandon AI.
English
12
9
76
10.8K
Can Vardar
Can Vardar@icanvardar·
prove me you're not an ai
English
1K
4
517
63.5K
Jon Hernandez
Jon Hernandez@JonhernandezIA·
📁 Sam Altman, CEO of OpenAI, says AI has already crossed a line many thought was impossible. It does not just predict words. It is starting to discover knowledge that existed nowhere. The paradox is brutal, we trained a machine to imitate the past, and it began pushing the future forward. The limit is no longer memory. It is reason.
English
27
15
205
21.6K
Arpad Toth
Arpad Toth@arpadietoth·
@relizarov Because that way the generate more tokens and that means more money for companies :)
English
0
0
1
153
Roman Elizarov
Roman Elizarov@relizarov·
Why do LLMs use such varied, almost poetic language, inventing synonyms and piling on adjectives while doing otherwise mundane technical work? Is this urge to sound erudite a side effect of RLHF, or something else entirely?
English
13
0
35
6.2K
Arpad Toth
Arpad Toth@arpadietoth·
@kerckhove_ts The compilers are deterministic in how the bytecode will behave unles they have a bug.. LLMs are not.. from a vague context dependent english they woull spit out some code that might do VERY different things for similar prompts. Compilers are deterministic in this regard.
English
0
0
19
1.2K
Arpad Toth
Arpad Toth@arpadietoth·
@woctordho @pobocks @IceSolst 1. read Chomsky hierarchy to understand the difference between English and a programming language as input. 2. Even with zero temperature is still not fully deterministic, you can read about how tokens are sampled etc... is too much for a X reply
English
0
0
0
17
woctordho
woctordho@woctordho·
@arpadietoth @pobocks @IceSolst That's the point - how do you define 'enforce precision'? Even if it's a pure function, it's hard to see what invariants are kept under this function. Also, nowadays we don't deliberately randomize an LLM. A common usage is temperature = 0 for coding and 1 for creative writing.
English
1
0
1
32
Arpad Toth
Arpad Toth@arpadietoth·
@woctordho @pobocks @IceSolst No I mean that the input of an LLM does not enforce precision, like a context-free grammar would. Also the ouput of an LLM si deliberately randomised.
English
1
0
0
19
woctordho
woctordho@woctordho·
@arpadietoth @pobocks @IceSolst The tokenizer of LLM is context-free. I guess what you meant is 'pure function'. Both the compiler and the LLM can be seen as a pure function of the input.
English
1
0
1
28
Arpad Toth
Arpad Toth@arpadietoth·
@woctordho @pobocks @IceSolst Let me phrase it differently: the compiler has also an input that allows deterministic output (a context-free language), but LLMs besides the randomness in their ouput, have also a context-dependent language as input, so in practice they are not.
English
1
0
0
21
woctordho
woctordho@woctordho·
@arpadietoth @pobocks @IceSolst The interesting thing is that LLMs are deterministic mathematically. The DeepSeek V4 tech report spent a lot to explain how they make their pipeline entirely deterministic. But the implementations of compilers and LLMs are another question.
English
1
0
1
35
Arpad Toth
Arpad Toth@arpadietoth·
@IceSolst @gerardsans So instead of writing deterministic code, you put a statistics machine to generate some random garbage and then WRITE MORE CODE to verifiy that the random garbage is good, instead of writing the needed code in the first place... clever..
English
0
0
0
18
solst/ICE of Astarte
solst/ICE of Astarte@IceSolst·
@gerardsans Yes that’s the point here: input to a compiler is untrusted, and a compiler is a deterministic program to validate certain attributes of that input Which is similar to what we need for LLM output
English
1
0
2
953
Arpad Toth
Arpad Toth@arpadietoth·
@IceSolst @LittleKeegs0 that tool suite will be non deterministic also if you write it with AI. If you write it manually then why not write the code directly?
English
0
0
2
75
solst/ICE of Astarte
solst/ICE of Astarte@IceSolst·
@LittleKeegs0 That’s the point: it’s because agents are nondeterministic that you need a tool suite to validate output correctness
English
4
0
4
2.3K
Arpad Toth
Arpad Toth@arpadietoth·
@IceSolst @harderthanfire It doesn't need to because programming languages are context-free, meaning a machine can reliably convert it into binary without loosing meaning. English is not context-free and cannot reliably parsed to binary. There is an entire science regarding this: arpytoth.ro/2026/05/02/its…
English
0
0
3
52
solst/ICE of Astarte
solst/ICE of Astarte@IceSolst·
@harderthanfire No the vast majority of devs are not reversing binaries. We design tooling for the average dev, who absolutely does not read compiler output.
English
13
1
43
7.5K
Arpad Toth
Arpad Toth@arpadietoth·
@woctordho @pobocks @IceSolst Math, you can prove it mathematically. This is exactly how you know that LLMs aren't because you can prove mathematically they aren't.
English
1
0
1
20

Keşfet

@Zhuinden @hasen_95dx @KaiXCreator @unclebobmartin @plainionist @awarehossain @SumitM_X @vlad_mihalcea