“… sometimes the coding agent gets things wrong, but that is revealed immediately…”

I’m curious about the belief all coding errors show up immediately. I propose the situation is more similar to fields like architecture, construction, medicine, etc. where catastrophic problem can hide for weeks.

Thinking about this more… I guess if the alternative to “free market” is “regulated market”, I agree with you because the shortage was not provoked by the government. However, I think I am pushing back on drawing conclusions based on potentially-unsustainable conditions. (ex: sub-prime mortgages)

AI is largely considered to be a bubble that cannot (currently) sustain itself. Shortages driven by irrational/unsustainable behavior do not feel like they should be called free market forces to me.

“The harsh truth is that the free market has decided that the components that go into consumer tech are so valuable…”

Conventional wisdom is that consumer tech components – like memory and storage – have skyrocketed in price exclusively due to AI buildout. Premature to call that “free market”, IMO.

At some point after the release of 5, they broke the behavior of my wheel. That wheel still works great in 4, but 5 was unplayable even months later when I checked for a patch (XBOX, BTW). IIRC, the sensitivity was cranked to 11, so precision was impossible. Bummer because I love the Forza series!

As an experiment, could you use Task Manager to give high priority affinity to the running process and see if it behaves better? Or prove that scheduling is NOT the problem?

I think I read that on Mac/iOS scheduling to power/efficiency cores is based on process/thread priority level. Maybe something like that?

Fair point, though I think it glosses over something significant: error rate. Reputable newspapers (or blogs like your own!) publish retractions when they get a fact wrong. But that’s a rare event. LLMs hallucinate by design and that’s a common event. The “trustability level” is wildly different.

Not so easy if you want them to preserve tab stops, perhaps. :)

Interesting! In the meantime, I had realized a simple way to improve this with just a tiny bit of CSS. I made a note to look at that when I get a chance.

I remembered why I didn’t do that in the first place! Old posts are fully HTML vs. newer posts which are Markdown and pass code blocks through a highlighter during render. There’s no render step for old posts to give an opportunity to reformat those samples. And static edits wouldn’t be theme-aware.

Context: dlaa.me/blog/post/81...

Sorry! Older posts use hard-coded HTML for light theme. That shouldn’t be the case with newer posts which use highlight.js. For example: dlaa.me/blog/post/ic...

I’ve made myself a note to see if it’s safe to remove the formatting from old stuff and let the auto-highlight code handle colorizing.

Eww, sorry.

Only Administrators members should be able to write there, so it seems like they run their service as Admin OR they relax privileges. Both are anti-patterns, I think.

Are you getting Beetlejuice vibes? Because I’m getting Beetlejuice vibes.