CraigJPerry
Are you of the view that LLMs have more potential than the internet? The internet has been around for so long that it's had time to have actual impact and yet it barely shows up in GDP. It still has so much untapped potential even yet. LLMs are too new to have amassed much impact yet, certainly not on the scale of the internet. But they have lots of potential still untapped. I just don't see how we get to even 5% GDP impact with them (either positive or negative). But i could be in a blind spot, genuinely curious to learn another view point.
[1] I've built all sorts with them. I've built various ML models but the only LLM i've built was effectively following a tutorial, nanochat
A clear weakness is that mileage readings can be altered. This used to matter on a smaller scale: dodgy second-hand sales, but now more or less anyone could be considered as potentially incentivised to interfere.
The mileage must remain a changeable field since the recording component may need replacement, for example, after accident damage or to correct a fault in the assembly that recording memory is part of.
The only reason i couldn't tell memphis is because it's aping a known design language already, it's just a perfect copy so me not being able to tell it wasn't AI isn't a free pass in this case.
I think if there was a little less "grid" feeling to the layout details then i wouldn't be able to tell on a few of the other designs either.
Edit: the designers republic version is quite a lot different than the others, it's not just that it's the only design to avoid pairing every feature with an icon, it's the whole layout variety - there's different font sizes, the positioning of elements is very well done to my eye. I'm gonna say this has to be another case like memphis, there's some pre-existing design language that i don't know of and it's just aping that. It's much more human than the others which i can believe are straight ai.
In test cases i can do anything, a test framework is just a way of discovering and then scheduling functions to run. I can emit useful instructions to the agent from the failed test case: "After walking the AST of all use of state machine X, a branch was found at Y which reused stale state. Ensure stale references are dropped..."
I can force the agent to pass the test suite before it considers itself done. I can reject edits of such test cases to partially mitigate reward hacking. etc etc
But I learned something really basic - i didn't know that the ANE (and the data pipeline around it) was designed for CNN rather than transformers. It's always been an open loop in my head, wondering why the ANE was less impactful than i understood it should be.
But once you've toppled the house of knowledge work, you'll find you can rebuild it easily because the foundation still exists, and it's the thing missing in this analysis; judgement.
I have switched to pnpm already.
I don't think it's common knowledge how much they dropped imports since the Iran / straight of hormuz crisis.
Back at the start of that conflict there were a lot of breathless takes about what catastrophe was in store in a few weeks time, and what materialised was bad but not in anything like the same league as what was predicted.
The thing that wasn't accounted for in the analysis at the time was just how rapidly and by how much China could suppress its appetite for oil imports.
System 1 can refer to the dashcam or the mk1 eyeball + notepad in your example.
System 2: a system for tracking the presence of a person across both time and location
An example of system 2 would be the facial id system being trialled on the london underground currently.
These are a difference in kind not in degree. It doesn't matter how many system 1's you deploy, you cannot unlock the capability of querying where any given face was observed across time and location.
Where did you learn that? It doesn't reflect the structural volumes present. Central banks make interventions all the time in line with little stabilisation programs. Those are almost always deemed success.
Maybe you deduced it by yourself? If so, fx is weird despite traditionally being seen as the simplest area in finance. E.g. It's counter-intuitive but we tend to think trade make up most FX volume globally. It's in the area of less than 3%. The majority by far is speculative and hedging.
The other trap is fx volume, people assume the know what volume is but then they learn expressions of fx volume is almost always actually tick volume.
Maybe i will write a blog post, if nothing else than to get my own thoughts in order.
I don't want to be overly critical, there's languages that people complain about and then there are languages that no one uses... If i compare it to js/ts, some stuff is genuinely better in Python - e.g. if you missed an await. While both ecosystems have lint tools available for this, but the behaviour is just friendlier in python.
Structured concurrency is better in python, but even if TaskGroup is nicer to use than AbortController, it still has its own foibles which means i'll usually advocate for AnyIO.
But the js/ts ecosystem just generally benefits from being async from the get-go where python you're just a time.sleep() away from a bug that will slip through dev envs and ci pipelines undetected and only rear its head under load in prod.
If i have a tip to share its the debug flag for asyncio run:
asyncio.run(f(), debug=True) # find some more issues before prodIt's not going to be the right stack for long lived websocket connections or whatever but for a CRUD-ish or enterprise app, often very productive.
[1] buffer bloat is too easy to sleep walk into
queue = asyncio.Queue() # oops
await asyncio.gather(*(fetch(item) for item in items)) # look mum! no outbound sockets left or maybe even no file descriptors
data = requests.get(url).json() # oops! we're blocking the event loop
I can go on and on about async python (cancellation bugs, leaking a task - that has a cataclysmic failure mode where if you forget to hold a reference to your asyncio.create_task() then it's all weak refs in that machinery so your task can get garbage collected before it ran or completed in production! super tricky forensics. Then there's all the obvious stuff like race conditions, new ways of creating deadlocks, blah blah i really can go on for days.
Does zig have auto vectorisation? I'm thinking that if you write the code in a vector friendly way, then the compiler can do the boiler plate for you
https://llvm.org/docs/Vectorizers.html
https://inside.java/2025/08/16/jvmls-hotspot-auto-vectorizat...
And From: https://getdx.com/research/measuring-developer-productivity-...
> Diffs per engineer*
>
> * Not at individual level
I found "initiatives" was added. What does this new initiatives measure bring. Why do we care about otherwise unqualified initiatives, how do i know that doesn't just mean using the other 4 proven measures as cover for pushing pet projects without merit?
I'm super cynical tonight it seems. This is just rubbing me up the wrong way i guess and i can't really put my finger on why.
Incidentally, Whenever i've done this in the past it's had a pleasant side effect of improving architecture. You end up forcing something akin to "push for's down and pull if's up" because crossing the ffi boundary is not free. It can be quite magical, as in leading to comically unbelievably speed ups when you also take advantage of vector intrinsics.
what if the market was hard to enter?
What if the inputs aren't available?
What if the costs of the inputs rises?
What if the capital for expansion isn't available?
What if the manufacturers don't expect demand to persist?
What if there's a shortage of skilled labour?
What if it takes years to expand supply?
What if there wasn't effective competition?
I'm not sure it's as certain as you seem to claim.
Could be an Alpina
This is a circa 1990s engine in the US market i think? Dual Overhead Cam didn't really become popular in the US market until then i think. 70s-80s for single overhead cam to become established.
The diagrams are beautiful and informative as always from this author.
There's no free lunch id think, the PG model is more robust. Unsafe extensions can take down the whole instance in the threaded model, processes contain the blast radius to that connection (also typically easier to debug since this type of issue is thankfully rare, it's also gnarly to get on top of).
Further, on linux (not on windows) a lot of the lines between a thread and a process get blurry (copy on write, shared memory mappings etc). They're both handled very similarly in the kernel, theyre both scheduled using similar machinery.
>> For PG to do plan caching it would need to serialize the plan between processes and that would require some significant work since it was never designed that way.
Is that true? I'm thinking the buffer cache and locks and WAL coordination are just as fast - it's just mmap'd SHM into each process. It's not like every access needs IPC?
Anyway i was going to post my favourite tool in this space https://github.com/jopohl/urh universal radio hacker just makes the process trivial, but i see the repo is marked archived now. Either way, the software is excellent.
But how do I get to express that demand? Asking as a frustrated regular user of excel - excel is amazing software but if your laptop is not in airplane mode, the number of little delays that creep in is wild. It's all seemingly network delays, connecting to onedrive servers when i'm editing a field (why?!), 10s of connections to random microsoft domains as i flick between tabs in the UI (why?!) - each flick incurring a subtle but observable delay.
>> Dreaming is free... All Electron devs
I like your sentiment for sure but i reckon you might be barking up the wrong tree. I'll give the clearest counter example i know of:
When i scroll a buffer in Zed (it's a 120fps editor written in rust that i really want to like) i perceive micro stutters.
When i scroll a buffer in VSCode (an electron app) it's buttery smooth.
I've tried this many times over 1.5+ years of releases. It's a reliable finding on an m1 macbook pro and an m1 imac.
If the slow stack can be fast and the fast stack can be slow, then there's more to this than just tech stack.
If something is judgement heavy, "code i care deeply about", then i don't really agree with the direction of travel here. Don't try to delegate decisions you care deeply about.
I do like the framing of agent loop vs harness loop, but only delegate stuff that you can accurately specify in advance, that usually means stuff that's repeatable in my case ("hey go see how i did X, do that but for Y"), and that inherently means stuff that's predictable.
For stuff where lack of my judgement as input is just going to cause me to say "no", we're down to collaborating in the "agent loop" as Armin puts it. And that's totally fine. It's fast, but also safe.
Remember before AI coding assistants, sometimes you'd get an engineer join your team who was SUPER productive, your peers would be jealous "oh yeah but you guys only got all that done because you have X on your team!" - they didn't live the curse of having that kind of person around - if you don't have them PERFECTLY aligned, then they run off at break neck speed in the wrong direction.
I'm more interested right now in what does that abstraction look like for AI generated code. Is there some reasonable solution wherein a sandboxed component in the enterprise architecture has various attributes (e.g. the bytes i stuff into this file store component are always the exact bytes i get back from it) confirmed by methods other than a human reading its code? Those methods, are they cheaper, faster, safer than just having a human do it?
If your enterprise architects have to read every line of code in your system today then i'd claim your architecture practices have room to mature. What can derived from that, and in which scenarios, for the purposes of safely leveraging immutable write-only code? I'm not interested in evolving the code (lines of code spent to solve a business problem was never an asset, it was always a cost) if it wasn't hand crafted by a human, i still have the requirements so i can just regenerate the entire thing with the revised requirement.
Author is top notch in my book. I'm a sucker for someone taking a complex problem and distilling out a simple solution. I don't know of higher praise to give a developer.
I have a GTX1080ti which i think is circa 2018, it's unused, more than paid for itself over the years, owes me nothing at this point so the hardware is free.
It runs Gemma e4b multimodal, qwen 3.5 8b or the qwen 4b embeddings models well enough (40+ t/s for the LLMs).
The machine consumes 350 watts at the wall when under load (3 watts when sleeping, 80w at idle). Electricity costs me £0.035GBP/kwh which is cheap for the UK (load shifting via house battery).
144k output tokens for around 1pence (and takes an hour to do that in theory).
It's only JUST cheaper to use than the far more capable deepseek v4 flash model despite the free hardware and ~10x cheaper than normal electricity.
By quirk of fate i've spent the past 2 days prototyping some stuff on pdfjs. Just trying to figure out a game plan for handling bounding boxes in the face of page zooming, different resolutions etc. etc. I can't see it mentioned whether the components are virtualising pages (as in reusing dom elements as document pages scroll by). I guess i just learned what i'll be exploring tomorrow then...