My most played Steam game is Dishonored, at 127 hours. I have replayed it a lot. A rarity for me, but I really liked that game. Dishonored came out in 2012. It’s taken me 12 years to accumulate that many hours.

Balatro came out two months ago.

I have 93 hours in it.

I used to work summers as an apprentice electrician. The amount of crazy wiring I saw in old houses was (heh) shocking. Sometimes it was just that it was old. Real old houses sometimes just had bare wire wrapped in silk. … And a few decades later that silk was frayed and crumbling in the walls and needed replacing.

My current house was wired at a time when copper was more precious, so it was wired up and down through the house, with circuits arranged by proximity, not necessarily logic. When a certain circuit in my house blows the breaker, my TV, PC and one wall of the master bedroom all lose power. The TV and PC are not in the same room either.

Just saying.

… Saying what, exactly?

I said that we should

• design for change
• “within reason”
• because we can’t know what exact changes are needed.

And you argued… The same thing? Just in the reverse order?

Have you ever been in an old house? Not old, like, on the Historic Register, well-preserved, rich bastard “old house”. Just a house that has been around awhile. A place that has seen a lot of living.

You’ll find light switches that don’t connect to anything; artwork hiding holes in the walls; sometimes walls have been added or removed and the floors no longer match.

Any construction that gets used, must change as needs change. Be it a house or a city or a program, these evolutions of need inevitably introduce complexity and flaws that are large enough to annoy, but small enough to ignore. Over time those issues accumulate until they reach a crisis point. Houses get remodeled or torn down, cities build or remove highways, and programs get refactored or replaced.

You can and should design for change, within reason, because all successful programs will need to change in ways you cannot predict. But the fact that a system eventually becomes complex and flawed is not due to engineering failures - it is inherent in the nature of changing systems.

… And can you fix it?

That sounds amazing - OMW to check it out!

Oh, for sure. I focused on ML in college. My first job was actually coding self-driving vehicles for open-pit copper mining operations! (I taught gigantic earth tillers to execute 3-point turns.)

I’m not in that space anymore, but I do get how LLMs work. Philosophically, I’m inclined to believe that the statistical model encoded in an LLM does model a sort of intelligence. Certainly not consciousness - LLMs don’t have any mechanism I’d accept as agency or any sort of internal “mind” state. But I also think that the common description of “supercharged autocorrect” is overreductive. Useful as rhetorical counter to the hype cycle, but just as misleading in its own way.

I’ve been playing with chatbots of varying complexity since the 1990s. LLMs are frankly a quantum leap forward. Even GPT-2 was pretty much useless compared to modern models.

All that said… All these models are trained on the best - but mostly worst - data the world has to offer… And if you average a handful of textbooks with an internet-full of self-confident blowhards (like me) - it’s not too surprising that today’s LLMs are all… kinda mid compared to an actual human.

But if you compare the performance of an LLM to the state of the art in natural language comprehension and response… It’s not even close. Going from a suite of single-focus programs, each using keyword recognition and word stem-based parsing to guess what the user wants (Try asking Alexa to “Play ‘Records’ by Weezer” sometime - it can’t because of the keyword collision), to a single program that can respond intelligibly to pretty much any statement, with a limited - but nonzero - chance of getting things right…

This tech is raw and not really production ready, but I’m using a few LLMs in different contexts as assistants… And they work great.

Even though LLMs are not a good replacement for actual human skill - they’re fucking awesome. 😅

What I think is amazing about LLMs is that they are smart enough to be tricked. You can’t talk your way around a password prompt. You either know the password or you don’t.

But LLMs have enough of something intelligence-like that a moderately clever human can talk them into doing pretty much anything.

That’s a wild advancement in artificial intelligence. Something that a human can trick, with nothing more than natural language!

Now… Whether you ought to hand control of your platform over to a mathematical average of internet dialog… That’s another question.

Man - that’s wild. Thank you for coming though with a citation - I appreciate it!

a quick web search uses much less power/resources compared to AI inference

Do you have a source for that? Not that I’m doubting you, just curious. I read once that the internet infrastructure required to support a cellphone uses about the same amount of electricity as an average US home.

Thinking about it, I know that LeGoog has yuge data centers to support its search engine. A simple web search is going to hit their massive distributed DB to return answers in subsecond time. Whereas running an LLM (NOT training one, which is admittedly cuckoo bananas energy intensive) would be executed on a single GPU, albeit a hefty one.

So on one hand you’ll have a query hitting multiple (comparatively) lightweight machines to lookup results - and all the networking gear between. One the other, a beefy single-GPU machine.

(All of this is from the perspective of handling a single request, of course. I’m not suggesting that Wikipedia would run this service on only one machine.)

This looks less like the LLM is making a claim so much as using an LLM to generate a search query and then read through the results in order to find anything that might relate to the section being searched.

It leans into the things LLMs are pretty good at (summarizing natural language; constructing queries according to a given pattern; checking through text for content that matches semantically instead of literally) and links directly to a source instead of leaning on the thing that LLMs only pretend to be good at (synthesizing answers).

In order to add their names to your dictionary. You don’t have to allow it. But given that there’s no internet access for the keyboard - it seems pretty safe

Thank you for responding! I really liked this bit

with a (decently designed) UI, you merely have to remember the path you took to get to wherever you want to go, what buttons to press, what mouse movements to execute.

I think that’s very insightful. I certainly have developed muscle-memory for many of my most-frequent commands in the CLI or editor of choice.

I agree about Visual Studio as a preference. I’ve used (or at least tried) dozens of IDE setups down the years from vi/emacs to JetBrains/VS to more esoteric things like Code Bubbles. I’ve found my personal happy place but I’d never tell someone else their way of working was wrong.

(Except for emacs devs. (Excepting again evil-mode emacs devs - who are merely confused and are approaching the light.)) ;)

I hope you take this in good humor and at least consider a TUI for your next project.

Absolutely. I see what you did there… 😉

But seriously, thank you for your response!

I think your comment about GUIs being better at displaying the current state and context was very insightful. Most CLI work I do is generally about composing a pipeline and shoving some sort of data through it. As a class of work, that’s a common task, but certainly not the only thing I do with my PC.

Multistage operations like, say, Bluetooth pairing I definitely prefer to use the GUI for. I think it is partially because of the state tracking inherent in the process.

Thanks again!

As someone who genuinely loves the command line - I’d like to know more about your perspective. (Genuinely. I solemnly swear not to try to convince you of my perspective.)

What about GUIs appeals to you over a command line?

I like the CLI because it feels like a conversation with the computer. I explain what I want, combining commands as necessary, and the machine responds.

With GUIs I feel like I’m always relearning tools. Even something as straightforward as ‘find and replace’ has different keyboard shortcuts in most of the text-editing apps I use - and regex support is spotty.

Not to say that I think the terminal is best for all things. I do use an IDE and windowing environments. Just that - when there are CLI tools I tend to prefer them over an equivalent GUI tool.

Anyway, I’m interested to hear your perspective- what about GUIs works better for you? What about the CLI is failing you?

Thank you!

Lots of little quality of life things. For instance, in Kotlin types can be marked nullable or not. When you are passing a potential null into a non-nullable argument, the compiler raises an error.

But if you had already checked earlier in scope whether or not the value was null, the compiler remembers that the value is guaranteed not to be null and won’t blow up.

Same for other typechecks. Once you have asserted that a value is a given type, you don’t need to cast it everywhere else. The compiler will remember.

Let’s start a patent troll company that exclusively deals in dark pattern bullshit. Then sue every company that implements any of our terrible patents for as much money as possible. Use the proceeds to bribe lobby congress to pass stronger consumer protection laws.

Kotlin is Java with all the suck taken out.

It’s a modern, ergonomic language that runs on the JVM and removes as much GD boilerplate as it can.

It’s fantastic.

That one dude still using Delphi is getting screwed.

Also, these salary numbers seem… real low. I get that it’s the median so maybe a huge number of overseas engineers are pulling the results down but in my neck of the woods 105K is less than what we pay juniors.

Argh. I hate that argument.

Yes - “Rewriting history” is a Bad Thing - but o argue that’s only on ‘main’ (or other shared branches). You should (IMHO) absolutely rewrite your local history pre-push for exactly the reasons you state.

If you rewrite main’s history and force your changes everybody else is gonna have conflicts. Also - history is important for certain debugging and investigation. Don’t be that guy.

Before you push though… rebasing your work to be easily digestible and have a single(ish) focus per commit is so helpful.

• review is easier since concerns aren’t mixed
• If a commit needs to be reverted it limits the collateral damage
• history is easier to follow because the commits tell a story

I use a stacked commit tool to help automate rebasing on upstream commits, but you can do it all with git pretty easily.

Anyway. Good on you; Keep the faith; etc etc. :)