Fr0styMatt88

What kind of speed in tk/s do you get with the MacBook?

Your see this in coding agents too. The only times so far I’ve really seen Opus tie itself into a knot is where I’ve asked it to fix something that I thought was broken but actually wasn’t in the way I had described. It will bias towards your description (I’m guessing because that’s the most recent context it has?).

Actually, I would be really pleased if a member of Parliament asked that. That shows a level of deeper consideration.

Isn’t there a saying about there being no stupid questions, only stupid answers or something?

I think it’s the combined depth AND breadth of knowledge that can be captured by AI models that is going to make them way better than most humans at this kind of stuff.

Exactly. I was about to reply to the comment with “perfect example of not knowing what you don’t know” in terms of self-diagnosis.

My internal model is/was “if the scan wasn’t set up / can’t detect the thing, why would the statement be present at all?”.

That implicit assumption is really subtle.

I guess ‘per GB’ doesn’t really capture it, because the base number of gigabytes available to people (ie- the smallest compatible RAM kit you need to build a computer) and the base number of gigabytes you really need (OS bloat, feeling responsive, etc) have gone up so much.

Were we really paying more for RAM per GB in the late 2010s than we are now?

Just really doesn't feel like it. Interesting.

Local using Qwen3.6-27B; 2xRTX 5070Ti graphics cards; VS Code with Cline at the moment and Ollama back-end (will get to trying the others soon).

It’ll be interesting.

I’m using Qwen3.6:27B at home and mostly Sonnet/Opus (depending on the complexity of the task) at work.

You have to break things down into smaller chunks for the local models. For the bigger cloud ones they can do a lot of the broader thinking.

If we can agree that the AI model is at least as capable as a junior engineer or new contractor, how’s that different to saying “software engineering isn’t worth $200 a month”?

Has a very race-to-the-bottom feel to it.

Though in the grand scheme of it, $200/mo probably isn’t the real price either. Also looking at it not just in a vacuum - paying for a product that can change what you get from under you doesn’t seem great anyway.

At least with a locally-hosted model you know what you’re getting.

It’s interesting — my own impression kinda matches that; I’m still doing most of the thinking when using AI that I’d do before writing the code, it’s just that now the feedback loop is so much faster.

One of the artists at my company described generative AI as “I get none of the fun of actually drawing the thing and all of the work of fixing it so it works properly with what we need”. That’s not how I feel about it in relation to writing code.

Maybe I’m just tired and AI has made things interesting again.

How is the place you’re at approaching AI in this context?

As a senior I worry about the juniors coming in — Claude can do what I would have previously tasked to a junior.

I guess the shape of the junior role just changes.

As a senior you can get into a bad habit of being scared to make changes. It happens after enough experience with enough codebases.

It’s good to not just go change things for the sake of it — it’s equally as important to ask yourself if you’ve gone too far in the other direction and to always remain curious and critical of yourself.

I mean I can see it. Books are just a way to transmit one’s thoughts and experiences to other people. So it’s no different to being exposed to someone with a different viewpoint. Common sense isn’t common or innate, it’s tribal knowledge.

There’s that XKCD about someone learning something new that was just thought to be something everyone knew.

Also you don’t know what you don’t know.

Agree though — coaching and persuasion are a huge part which is why I think a lot of these books seem ‘fluffy’ if all you’re wanting is a collection of facts.

There was also the very early leak/theft of the Half-Life 2 beta source (https://combineoverwiki.net/wiki/Half-Life_2_leak)

We aren’t even being given a choice nor being consulted for our judgement.

Higher-ups are literally saying “This company WILL use AI, you are required to use AI, now go use it whether you want to or not”.

I wonder if this will change now with the model cost explosions that are happening.

I generally agree with you, except for one thing.

Sure, users don’t care if the code is a mess. At least at first. I mean, they can’t know.

But they DO care when the same bugs exist for years, when new features dry up, when it drains their battery, etc. When technical debt becomes due, users notice in other ways.

Now, whether they care enough or just don’t have a choice is another matter.

So I’m wondering what ‘rasterizing’ literally means in this case. I imagine it’s just creating a 2D map of elements at a very low (probably character) resolution, then diffing that against the last generated map to come up with an optimal ANSI sequence to send to the terminal, would that be right?

Seems like a cool puzzle to solve. I wonder what the engineering and organisation tradeoffs were that lead to it — does it let them reuse a bunch of existing code?

I wrote a TUI library back in the day for Turbo Pascal — it was essentially taking an immediate-mode approach (which in this context is just a fancy way of saying it was procedural haha).

Also fun, Google purchased movies over the years getting moved to Youtube and then seemingly losing all surround sound content and coming through as 2-channel only (Android TV).

Had a few eye surgeries (vitrectomies) under sedation, no pain but lots of flashing lights and lots of kaleidoscope patterns. It was pretty wild.

I was lucky that coming out of sedation was actually fantastic, like the only time I can remember feeling that blissfully relaxed was in maybe a few beach holidays I went on as a kid.

General anaesthetic scares me way more.

I actually think it would have done well if it was just like those button boxes / Stream Deck / etc. Like a row of transparent function keys with screens, but then that would have been a flexibility tradeoff.

Even at the time I remember it was widely cited that the SIM eject tool was a test for their new manufacturing process.

What I’m curious about is what lead to larks ‘winning’ in the sense that there’s this massive prejudice against night owls.

Though I have heard that there are natural biological functions that depend on the sun such that night owls who are sleeping their natural pattern are STILL predisposed more towards certain physical/mental conditions. Though who knows?

I _wish_ I could nap after hearing all the benefits, but for me it’s either doze off for five minutes and wake up feeling blah, or lie there for 10 minutes, MAYBE go to sleep for a bit and wake up feeling horrible.

If I have light to anchor my circadian rhythm, I’m happiest waking up around 5:30-6:00 and going straight through, starting to wind down at 8:30.

If I sleep later, I’ll end up shifting more towards naturally waking up around 10:30, going to bed at 11:30 PM and generally feeling not terrible but not great and slightly tired during the entire day.

Luckily the light can be artificial that wakes me up — I use smart bulbs as an alarm.

Have you tried the screen zoom in accessibility settings? The responsiveness is great with the trackpad gestures.

From a quick Google that kinda makes sense, it’s just the strong, _sustained_ power draw that gives them issues. So I’d say fundamental AND inverter design — imagine pushing 2kW continuously through an inverter.

It’s funny, power use can be really unintuitive. Try convincing someone that using the big air conditioner for heating is more efficient than using that little plug-in bar heater. Or yeah, a power board with 8 tiny wattage wall-warts isn’t using a lot of power.

I could probably run my big fridge overnight off my portable battery generator, but it wouldn’t run my small electric kettle without putting it into a special mode and for nowhere near as long.

Yeah it’s when you go off the happy path that it gets difficult. Like there’s a weird behaviour in your vibe-coded app that you don’t quite know how to describe succinctly and you end up in some back-and-forth.

But man AI is phenomenal for getting stuff out of your head and working quick.

I feel the exact same way about tutorials in games that try and be comprehensive and show you everything.

Incremental games do an amazing job at this (things like Universal Paperclips, A Dark Room, etc); parts of the game are revealed to you as you need them and it's often a fun surprise. I don't think the same thing is directly applicable to productivity apps, but I wonder if something could be taken from the pattern.

This is timely -- I'm coding an app at the moment and had the fleeting thought that "hey I should do a new user onboarding tour thingy" and then remembered that in general I skip them, so I havne't made one :)

Thank you, I was starting to wonder.

I guess because I’m in game dev maybe, but in all my jobs knowing about the underlying stack has either been necessary knowledge or highly regarded.

I can’t think of any time in my career where knowing about the internals of the stack was ever frowned upon or where it’s been anything other than an advantage (especially when hunting bugs). I must have been lucky.

How did it get in? Isn’t Linus known for being rightfully fussy about what makes it into the kernel?

Would be an interesting story.