I can't believe that anybody would give an LLM access to production.
Always always always assume that your LLM will destroy your production if given the possibility.
In other words, always assume the gun is loaded.
2,848 karma · joined February 27, 2013
Always always always assume that your LLM will destroy your production if given the possibility.
In other words, always assume the gun is loaded.
I thought so too, spend a bit of time getting deeply into it, and then realised that nope it really isn't that interesting if all you care about is writing good software.
There are people in the Haskell community who is really into math and how it can be mapped to Haskell. And hey that's all good and fine but there are much better things to study if what you focus on is writing good software.
I would (for example) recommend learning LEAN. It is mind blowingly good for your brain and will make you a better software developer.
LLVM at the IR level has no understanding of the semantics of your language and therefore can't do the kind of optimisations that will really make a big speed difference.
I ended up having two custom IR's for a very high performance compiler I maintain at work. It made a big difference.
I spend a significant amount of time improving performance for our paying customers because performance absolutely matters to them. They might use a single simple feature hundreds of times during a normal workday. When we make that experience even 10% faster or smoother, they notice immediately and are genuinely delighted by it. That creates an enormous amount of goodwill.
And sometimes we can go even further and completely automate tedious workflows they hate doing manually. That’s even more valuable.
Developers with experience from the games industry understand this instinctively. Even when the software has nothing to do with games, they tend to have a deep appreciation for responsiveness, efficiency, and user experience. That’s why I’m always excited to hire people with that background.
Instead, simply be a professional and fix what is needed, while working on non-developer visible tasks.
Don't allow people with zero clue to make those decisions. You are the expect. Make the decision.
Some of the worst programmers I have ever worked with had 30+ years of experience. They basically spend all of their time fixing bug after bug in a never ending cycle because the software they produced was so fragile that it would crash if you just looked at it wrong or the temperature in the room wasn't perfect.
While others with the same number of years of experience had massive systems in production for years with not a single bug reported by the happy users.
I have tried everything from generating a complete, detailed spec using AI and then one-shot generating the code, to generating code one step at a time with me reviewing each result.
The speed is pretty much the same. But generating code one step at a time is IMHO vastly superior because I deeply understand the code and can easily fix issues that the AI get stuck trying to fix.
However if you make your own programming language and want to post about it on HN, please add something new to the language that makes the language interesting.
I don't think we need yet another look-mum-I-made-a-compiler-for-a-language-that-is-similar-to-N-other-languages-but-with-a-slightly-different-syntax-and-zero-libraries post on HN.
Just saying.
Agreeing to drop bombs on people far away, without being in any danger yourself, is not being brave.
Often is not goo enough. I always carefully review AI generated code and sometimes find and fix bugs.
Apparently (according to a Google survey I think?) AI's produce 1.7 times the number of bugs that a human typically does.
It felt like visiting a third world country.
My guess is that it is primarily a culture issue.
It is fun and IMHO the best way to learn LISP.
If you get stuck read through one of the many examples online and then try again without looking at those examples.
That said, it still frequently introduces subtle bugs, so I have to review every change carefully.
The real trick is learning when to use it. Some tasks are much faster to do myself, while others are faster with Claude Code.
I am forced to use Linux at work and I dread updating anything because the result is usually a broken system. I never have that problem with Windows.
So YMMV is as always valid here.
In other words, yes AI's could in principle have consciousness.
However my #1 productivity tool is still a custom code generator I have been using for years. It routinely generates 90+% of the code needed to write a typical biz web application, leaving just the business logic.
No AI. Just straightforward high-level-spec-to-server-client-DB code that is 100% trusted and proven in battle.
Developing a "new kind of programming system" has been done many times before. Think case tools, functional programming, object oriented programming, CORBA, DDD, etc. etc. etc.
So best of luck but I highly recommend not investing your time unless you have at least one truly new idea that will change the outcome for the better.