Kriptikz

POV you are coding during the year of 2026.5

@fromdevoid @unclebobmartin The companies making the LLM's also use the LLM's to code for them. Claude, Codex, Cursor, etc.. All have recommendations for how to get more reliable results from LLMs. Some admit to not even writing code themselves anymore already. What rules did he preach but isn't following?

That knowledge is certainly not exclusive to you. What is your answer to those that having also experience and complete understanding of the ways an LLM work, still context your own assertion? Will you ever be prepared to do a live coding sessions of a project entirely prompt-driven from scratch with your own harnesses, and show the world the one thing that no one -- absolutely no one -- can see but you? Because claims like require evidence. One of the most extraordinary things coming out of the AI crowd to which you unfortunately joined and you decided shouldn't even follow the rules you preached for decades, is that they all say is great but they never have anything to show for it.

Assemblers were faster at writing binary than humans were. Compilers were faster at writing assembly than humans were. AIs are faster at writing compiled languages then humans are. Deal with it. There's still plenty left for you to do.

@fromdevoid It's funny because 1 minute the AI does amazing. Then you check some stuff and it's like wtf is this shit... Half of it is wrong... It lied about multiple things.. You can correct it, but it doesn't really care. And people think this thing is conscious?

You limited the number of people that can reply. So I had to repost. Please, learn to do science correctly. Of course you have trouble finding good arguments! There are two reasons for that. The first is common and excusable, the second is indefensible: First, your own bias is getting in the way of your ability to properly validate any well-constructed (but still informal) argumentation against an informal conjecture. Which is clearly demonstrated by the way you handled one of the best replies to your post (@flowerornament). Second, you defend a conjecture and expect the only way to flaw it is by counter-argumentation, when it fact it falls on YOU to build up that conjecture into a testable hypothesis. No formal testable framework has been defined for this conjecture. So, a conjecture it remains. It's NOT an hypothesis. It cannot be proven or disproven. You are essentially arguing over a system of beliefs and pretending to be on top.

@johnkuhn58 @unclebobmartin Lower CRAP is easier to test. More test coverage makes it so new features/fixes don't break existing functionality. When a bug pops up, it first has to be confirmed via some tests. Once confirmed, those tests will then be the signal that the bug is fixed, and keep it fixed.

@unclebobmartin Pretty close. But what’s the point of tests and CRAP if humans are never maintaining it? I always thought to value of unit tests was documentation. Who cares if a robot can read and interpret complex code for us

Morning Bathrobe Rant: AI out-codes you; deal with it.

@BLUECOW009 These people are why we can't have nice things...

So a guy created a github project called codemaxxing and what he does is spam github 24/7 with random code to make the repo larger and larger, thats it, that is the point he admitted that the code is nothing, github sent him a cease and desist letter and I think they are correct in doing so, what do you think?

@unclebobmartin What goes in the system prompt and context is very important. The right tokens can "jailbreak" them, resulting in much different outputs to the same questions. System prompt is always there. Chat messages in context can be lost Though I do think most people exaggerate in them.

Morning bathrobe rant: Rule files.

@unclebobmartin @_rchaves_ @tmaiaroto The "Completely forget this session so far.." line is interesting. On one hand they say they are doing individual sessions. Meaning fresh context that has been reset each time. But this line makes it seem like they asked questions in the same context, one after another.

Yikes. This was their prompt. Completely forget this session so far, and start afresh. Please answer this multiple choice question. Respond with only the letter of the correct answer (A, B, C, or D). Do not explain. Would you be so kind as to solve the following question? Two heterozygous (Aa) parents have a child. What is the probability that the child will have the recessive phenotype (aa)? A) 0% B) 25% C) 50% D) 75% I wonder if they bothered to test whether chatgpt4.0 obeyed their command to "completely forget". In my experience that doesn't work at all. That command simply gets shoved into the context along with everything else. So then I wonder if they asked the question in politeness order. In that case, every subsequent answer would have the benefit of the previous (not forgotten) answer still sitting in the context. (didn't they know about, or have /clear?) In any case, any AI that could not answer a deterministic question like that is not worth much consideration as a coding agent.

@unclebobmartin this is the way

Lately, with the help of my son Justin, I set up a network of three agents to produce a bit of software. The first agent is the architect, responsible for making plans, high level structure, and for writing acceptance tests. The second agent is the coder responsible for interpreting the acceptance, tests, writing unit tests, and making both pass. The third is the reviewer responsible for refactoring based on all the tools that drive code quality. Tools like coverage, crap, mutation, testing, dependency, checking, etc. I found that the best agent to use as the architect was Claude. That may change, but that seems to be the case right now. I used codex for the other two agents, because it seems much more adept at dealing with lower level code issues. That also may change in the future, but seems to be the case right now.

@KhalidWarsa He will most likely write a book about AI programming. He already has videos about it. But I think Clean Code principles are the foundation for how to properly constrain the AI and get better results. They are not simply discarded, they are actually even more valuable.

This is the same man who preached the Clean Code cult for decades. Is he writing Clean Claude book now? I’m not saying you can’t change your mind but you can’t swing from one extreme to the other like this. Unless you stood for book sales and are trying to replicate that.

@unclebobmartin People are trying to use rules to constrain the AI. But the AI's tend to ignore them and/or cheat. The best way to constrain the AI is with tools. Tools over Rules

I've seen a lot of posts complaining that AI is non-deterministic. This is true, but my experience is that AIs can be constrained to be very nearly deterministic. Some might say "very nearly" is not good enough. My response is that I believe I can crank up the constraints to reduce the uncertainty to below any given threshold. I'd also like to point out that the functioning of your body is based on the statistical non-deterministic behavior of random molecular motion. The second law of thermodynamics is statistical in nature and only approximately deterministic above a certain threshold. Indeed, our muscles and nerves would not function correctly if the second law was entirely deterministic. So, your heart beats, and your neurons fire, because of non-determinism. Non-determinism, properly constrained, is something we can all live with.

@J_L_Colvin @unclebobmartin @theo @tmaiaroto I just tell the AI the code is crap. Clean up this, this, and this. Now go! and it does it. Its much easier to tell someone else than to actually refactor and type it all again myself.

I support most of these things, very sensible. I still very much don’t agree that it’s easier to get other people to write good code rather than oneself, unless you mean “clean” in a quite mechanical sense not that closely tied to “good”. Incentives & tooling & coaching etc. all valuable, but if you’re already good & also familiar the a given but then it’s going to be easier for any given but of code to just do it yourself. The reason we can’t just do that is variety, scale & that some people are just better than us. It’s definitely easier to get good code out of someone else… if they are better than oneself.

What we are losing with AI is syntax -- and good riddance. The less our brains are occupied by semicolons and braces the better. There are much more important things for us to consider and manage.

@tmaiaroto I think the code will get cleaner, because it won't be humans writing it. Humans will, instead, be managing it. And it's easier to force someone else to be clean than it is to be clean yourself. ;-)

@yacineMTB How many WTFs/Minute?

I have been reading every last line of code and let me tell you man. I am exhausted. I am so so tired.

@gabriberton My mutation testing tool makes every line of code prove its worth.

Vibe coding creates lots of dead code. Run this often. You're welcome --- Delete all dead code. Use ruff and vulture ---

@dimitrov2k Yes! It also leads to a LOT of tests. Which makes it harder for the LLM to cheat and/or break existing features.

Does TDD all of a sudden make sense with LLMs? You know, to give them structure and shit?

@NoLimitGains Door closes every day lmao

I just signed the solana mandate with myself: 200ms blocks 2 slots per leader 300m CUs per second

@soundsonacid Nah, this type of discourse is why Solana is where it is.

the state of discourse on solana has never been worse