If the LLM can’t run the original code, it can’t compare to verify the translation is accurate

LLMs are surprising good at looking at plain old hex of a binary and rewriting it as source code that functions similarly.

I'm sure for these older games they'd succeed most of the time.

Dunno about that. I am on a TRS-80 CoCo(6809 CPU) group elsewhere and the oldschoolers are creating some impressive things using Claude and the like.

When the runtime is completely different, I'd also argue it's emulation more than translation? Anyway. Lost in translation I guess.

If you think other things are more important than writing, why would you expect anyone to take the time to read it ?

It’s just not a very efficient strategy, it might work for some readers, but as the comments show it’s a big turn off for many. Where it says “There’s one honest footnote”, that’s really grating, like the other commenter I also stopped reading there.

That’s not necessary rational, it’s just that we are social creatures that are really sensitive to those kind of surface level cues in communication, to gauge credibility. I’m sure many authors today use LLM in brainstorming, research, ask it to steelman an argument etc and no readers complain because it doesn’t impact the authorial voice.

The problem then is with letting LLM ‘fleshing out the details’: you end up having the LLM’s style and words all over your text. You can’t really do that without your authorial voice loosing credibility. I think next time, you have to go over it and edit it and make it sound like you.

If you don’t want to that, the good news is that I think people will start to get less critical of sloppy writing, as long as it sounds human. So you could sound convincing while leaving some details raw and not have the LLM have a go at it.

My approach is more "ship fast, iterate later". Your criticism of AI slop is valid and I share it. I've manually re-written the post today, no more AI slop. Of course, that came with the opportunity cost of not spending time on 100 other projects competing for my attention. AI has enabled me to do a lot more with my limited resource of attention.

I read this take on hacker news all the time, but what indication do you have that this would be popular ?

All the streaming giants had ample budget and engineering talent to make interactive fiction happen, but didn’t, presumably because they know a) not everybody is creative that way and b) a big part of culture is that you can discuss it with others, which doesn’t work for personalised works.

If you have your kids you’ll know that the shared culture we used to have is no more. Kids these days don’t turn on the tv when they get home from school but instead go on their highly personalised feeds… so for any shared culture, we’re now back to memes (in the original sense of the word, not internet memes).

You can’t look at the big giants with their budgets and think “If they can’t do it, nobody will”. The bigger you are, the slower you’ll move. This is exactly why FAANGS buy innovation rather than build it all themselves.

I’m 100% positive we’ll see exactly what I described in the coming years and it will be from an unknown startup with two founders, no budget, and just a coffee machine to keep them going

So any interesting alternative use for the Camlink ? All I can think of is dropping framerate (some of my cameras output 50p but are 25p). Or maybe adding a LUT ?

Since dieselgate I no longer believe in German engineering. That, and that they just can’t make their trains run on time. Never have I taken an ICE train that was not delayed. How could that happen in an actual engineering culture?