HN user

devbent

492 karma

https://meanderingthoughts.hashnode.dev

https://www.generativestorytelling.ai/

@thedevlinb

www.github.com/devlinb

meet.hn/city/us-Seattle

Socials:

- github.com/devlinb

- x.com/thedevlinb

---

Posts2
Comments206
View on HN

Find harder problems to solve.

I can easily make Claude freak out and run into limits. Claude is amazing but it only works at the abstraction level you ask of it, so if you ask it to write code to solve a problem it'll only solve that immediate problem, it doesn't have awareness of any larger refactorings or design improvements that could be made to improve what solution is even possible.

I'm building a simulated fantasy town backed by an LLM. Visitors to the website can vote on what the hero does next, and everyone in the town will react naturally to all of the heroes actions.

Unlike previous efforts in this space, the technique I am using consumes very little context, and I'm hoping to get it running on consumer GPUs.

UTF-8 wasn't a thing when the decision to go with UTF-16 was made.

UTF-8 became a thing shortly thereafter and everyone started laughing at MS for having followed the best industry standard that was available to them when they had to make a choice.

Windows has had multi-user support for ages.

That is how user switching in XP works, and how RDP works. You can have an arbitrary number of sessions of logged in users at once, only limited by the license for what version of Windows you have installed.

There have also been versions of Windows that allow multiple users to interact with each other at once, but I believe these have all been cancelled and I do not know to what extent these simultaneous users had their own accounts.

What the unholy fucking hell.

Where none of the people running these schools ever kids? Did they just magically appear from some portal from another, sad, apathetic, dimension? Or where they spawned out of eggs laid by depressed lizard people?

My school had 2 recesses, one 15 minute morning and one 30 minutes at lunch. They were absolutely crucial to social emotional development, and to just being outside.

6 hours of focused learning for 5 year olds? What type of massive idiocy is that. Ask the brightest college students how many hours of classes they sign up for, and unless you are talking to a crazy person, the answer is 4, with 3 being preferred by most (and considered a full course load!).

There is oodles of literature on how many hours a day a person can sit down and learn. 6 hours inside for a kid is wildly outside of recommendations.

Going Buildless 2 years ago

I am going build less for a recent project. The main difference for me is jsdoc vs typescript. There was a large initial learning curve, and the syntax is ugly, and having to duplicate all my TS types from the backend as JS doc types on the front end sucks, but at the end of the day it does work.

I was surprised how well module imports work in modern browsers.

I've never used css preprocessors before, so that wasn't an issue for me. I also have never nested CSS files, so that also wasn't an issue for me.

Another pain point is not having type safety on my front end calls to my backend. With TS I could use Zod on the front end as well and have type checks in place before I make those API calls. While obviously js doc tries to ensure at compile time that things are the correct shape, that isn't the same level of assurance.

The other major difference is that table top gaming is a constrained environment where one is allowed to act socially different than normal, and mistakes at the table are left at the table.

Players are allowed to play someone with a completely different personality, and spend extended amounts of time stepping into someone else's head and thinking how someone different would act.

There aren't any other situations in life where you can take someone socially awkward and say "ok now act really charismatic" and then, w/o judgement, that person spends 2 or 3 hours a week for months on end trying to figure out what a charismatic person would say and do.

That top logo is horrible and childish looking. The new logo will look dated soon, but at least it looks professional.

And 80k for a logo is incredibly cheap.

IMHO the problem is the majority of new multifamily construction is rentals only. Rentals ran by international firms suck the life out of communities, as people cannot be part of a local community when rent hikes can force them to move every couple of years.

We need large scale multifamily housing for sale.

I don’t get how one imagines being an animal and… it just exists as an automaton? And for some reason exactly humans are all conscious/intelligent,

So far evidence points to the simpler solution that all living creatures are automatons and that there is no mechanism by which self determinism can exist.

Wide acceptance of such a belief would pretty much ruin society so it is best if we all just go on pretending we are masters of our own destiny.

I'm working on something similar but instead of basing my efforts on the generator agents paper I'm basing it on a technique I call narrative generation. It requires far less context and fewer prompts, and focuses on generating narrative aspects and letting the traditional game engine things simulate the remainder of the world.

As an example, with the system I am building you only need to input what action the player has taken, and all the NPC's actions and dialogues will be generated.

I had an XML file format from one app I needed converted to a json file format for another app.

I threw both schemas at Claude and asked for it to write converter code.

Writing mocks, Claude saves an hour+ when mocking out complex classes.

I've never written graphics code before, I had a png animation film strip, Claude wrote code to load, parse, and animate it.

Genuinely curious if people think this is a product, solving a real problem?

For people with memory issues, products like this can be life changing.

On a second note, imagine a world where everyone has perfect memory recall. The concept of "recording" someone is useless, as everyone around a conversation (e.g. sitting at the next table at a cafe) would be a 100% reliable witness to the conversation.

Obviously we would not fault someone for taking steps to train their memory to be better, so why fault them for using electronics to improve the limits of biology?

One problem that I run into with LLM code generation on large projects is that at some point the LLM runs into a problem it just cannot fix no matter how it is prompted. This manifest in a number of ways. Sometimes it is by bouncing back and forth between two invalid solutions while other times it is bouncing back and forth fixing one issue and while breaking something else in another part of the code.

Another issue with complex projects is that llms will not tell you what you don't know. They will happily go about designing crappy code if you ask them for a crappy solution and they don't have the ability to recommend a better path forward unless explicitly prompted.

That said, I had Claude generate most of a tile-based 2D pixel art rendering engine[1] for me, but again, once things got complicated I had to go and start hand fixing the code because Claude was no longer able to make improvements.

I've seen these failure modes across multiple problem domains, from CSS (alternating between two broken css styles, neither came close to fixing the issue) to backend, to rendering code (trying to get character sprites correctly on the tiles)

[1] https://www.generativestorytelling.ai/town/index.html notice the tons of rendering artifacts. I've realized I'm going to need to rewrite a lot of how rendering happens to resolve them. Claude wrote 80% of the original code but by the time I'm done fixing everything maybe only 30% or so of Claude's code will remain.

without all the BS that comes with TypeScript, etc. JavaScript is so powerful and amazing, it’s a bummer to hamstring your app by using TypeScript and classic OO inheritance models

Typescript is no more OO than JS is. Typescript is simply a way to document, and optionally compile time enforce, what fields objects have and what objects methods/functions take in and return.

That is all TS does. It documents what is already there so you don't make typos on field names.

Lately I'm writing a project w/o typescript and I spent an hour finding a bug the boiled down to "tilewidth" vs "tileWidth". (The code didn't explode until much later when a computed value didn't exist)

That is the type of bug TS prevents from ever happening.

Auto adjusting other parts of the car based on what is happening, e.g. my car reduces fan speed when a call comes in.

Typically cars come with their own complete voice control system.

Customers expect much of the car to be accessible remotely and well, my GM Bolt EUV is far behind everyone else in this regard.

A few misbehaving CSS filters can make my discreet GPU turn on and at that point my battery life is a goner. Not sure who to blame in that scenario.

There was an old bug in FF around 2018 where a tab using a GPU would prevent a Windows laptop from ever sleeping. That ended up destroying that laptop's battery after it got thrown in my backpack and overheated a couple times.

epicenv set KEY VALUE -e myenv -p

I would have flipped this logic, default is local only and an extra param is needed to make them shared. variables as private should be the default, an extra switch should be needed to make them shared.

Aside from that, seems like a cool project. No Windows support is unfortunate, as I typically try to make my projects work across all 3 major dev platforms.

I'm not sure how this is better or worse than systems that use encrypted .env files, but one obvious disadvantage is for any but the most trivial use cases I end up having per environment configs, and this tools seems like it doesn't have support for multiple environments.

My newest open source project[1] is ~90% written by LLM, but I work in the field so there is a chance my ability to manipulate LLMs is above average.

It is also a tiny project, just a single microservice, although my upcoming open source project is fairly substantial and I suspect it'll be at least 30% LLM generated.

The fact is LLMs are really good at generating chunks of code, sometimes even entire files, but they aren't that good at making systematic changes to large swaths of existing code. They can be made to work for that scenario with tons of prompt engineering and prompt chaining, but that isn't where LLMs shine.

For example almost all of the HTML in that project was made by Claude. I then modified the code and re-uploaded it and asked Claude to add collapsible info boxes to parts of the page after the fact, and Claude completely failed at the task.

Still though I am seeing an incredible increase in my development velocity.

[1] https://www.github.com/devlinb/escaperoom -ugly UI, designed to help people create escape rooms for LLMs. A hosted version will be up later today!