zingar

It would be interesting to know what the guardrails are. That would help me with understanding how I can use AI content. For instance I asked Claude to help me draw a diagram to represent a software engineering concept for a public presentation and then I had to stop and think: am I about to just reuse something from a Martin Fowler or Kent Beck book without attribution?

I feel like I'm seeing a maths+AI change from "let's test the limits of LLMs by seeing if they can do useful math" to "LLMs can do useful math, now let's solve lots of problems!", or put a different way the goal has shifted from "interesting exercise for AI" to "making a big difference in math". Am I correct?

Are there practical applications of any these problems being solved? No judgement implied, I'm well aware that "no" only means "not yet".

Ah of course, I do have a blind spot with local models.

Now that you mention it, Claude has had trouble at times with balancing brackets in the elisp I have it write.

Yet I still had the idea that LLMs should be better at lisp than other languages.

A fascinating contradiction, thanks for pointing this out.

I haven’t seen an LLM generate syntactically invalid code in any language in … a year? So I don’t know if “more likely to be syntactically valid” is a good enough selling point.

But maybe I’m interpreting you too narrowly?

The 1500 line react component that my unsupervised agentic coding side project produced says that it can get worse than we have now.

org-gcal … bit of a pain getting the Google API config right but so far so good.

Where does “big model highly quantized” start getting worse than “smaller model less quantized”? Is there a general formula or is it just trial and error?

I fetch my google calendar every few minutes to get everything into my agenda. The API request is async, but writing to the calendar file and some additional processing made Emacs hang for 10 seconds at a time. The LLM-inspired solution was to spawn entire new emacs processes to do the work in the background. It's pretty cool that it works, but it feels like it shouldn't - and wouldn't with 2+ threads - have been that complicated.

This post is a breath of pithy fresh air. A small point made with few words that adds something worthwhile to the conversation.

On topic: maybe I live in a bubble but I'm surprised how many people on LinkedIn expose far behind the curve they are by talking about MCP or RAG or even prompt engineering as if it's a suddenly-new discovery.

This must the most compelling look I’ve seen at how software might work with LLMs doing a ton of heavy lifting.

There’s something kind of amazing here in that having read about property based testing I’m pretty confident I could apply it if I had a good use case.

This is great advice that AI (successfully) tempts me into ignoring.

What happened that made you switch away from contracting?

Yep, this was an issue long before Trump. They’ve just amped up the scale and stopped bothering with the deceit that they know doesn’t bother Americans.

To be clear these projects made it way further than they could have without AI. But any blocker that takes me out of the flow can easily become the thing that ends it. Not even necessarily because it’s a difficult problem, but just because I’m no longer getting that AI high.

The “old normal” described in this article sounds hellish. This sounds like all the worst practices that need to be drummed out of people who arrive not knowing what effective engineering looks like.

There’s no loss if all of this changes dramatically, there’s plenty of real work to do.

But honest question: where is this normal?

I have multiple side projects that I would never have contemplated building before but whose utility now exceeds the much lower cost to build.

I got a few weeks in to each and then stalled on all of them because the effort and motivation required to extend beyond the crazed early days _is_ still more than the utility I get.

In a professional context, paying someone for software to do something outside my core domain is still the practical option compared to the motivation and effort needed to maintain another dependency.

What’s the difference between a pattern and an abstraction?

The rugby player, absolutely not, I know the sport well. But interestingly this fictional player did play for my local team.

The neurologist I could believe exists somewhere in the world with the same last name.

I love a long form programming deep dive. I have questions, probably basic ones that belong in undergrad CS but it’s been … a while.

1. Can someone remind me why it was so important/intentional at the start of the language that every object has identity? 2. Why is it important that we not synchronize on these value objects?

I love the way this starts with using emacs badly since 2008. I start my own shorter emacs story similarly: “I’ve been using emacs for 10 years, so I’m only a beginner“.

Interesting. I started on spacemacs and never left. My spacemacs is super pimped for the things I want. It feels like if I rolled my own I would end up with something like spacemacs, but not as good.

So I guess you and others here have had the experience of building something that was your own that felt better than the distro?

Bahaha apparently only in their hallucinations. I’m not a professional rugby player or a neurologist.

It would matter a lot if there were 5 people reading/writing code or 100.

Is this “hacker news”-esque in terms of being a social bookmarking site? I don’t see much by way of the same topics, and don’t think the difference is only whether it’s Indy or not.

It’s reasonable to say that there are things like discovery and validation that are necessary for every startup (to succeed) and that some techniques for these can be automated.

That we can describe something like “validation” in the abstract and automate some part of it says nothing about whether it’s worthwhile. I’m hard pressed to think of anything that anyone pays for that doesn’t meet this description. Why should being a founder be special or different?

If this is true why am I hearing so much from “AI startups” about how they have big plans to hire?

100 feels low given I just saw dependabot do 8 in one hour. No AI required!

It matters a lot what size the PRs are, and this varies wildly from place to place. I spoke to someone who instituted a “no PRs over 500 lines” rule. I would refuse to even read something that big unless it was just a find and replace or boilerplate.

If I’m reading right you used to pay more for Claude but now deepseek has replaced that higher tier subscription. Do you mind my asking what you were paying before?

We just agreed on our team that we’re not posting AI-generated text into comms with humans.