@gloriouscow Yeah, it's hard to get right since the copy protection authors probed every corner for undocumented behavior.

@gloriouscow The biggest problem I see with AI isn't the technology but the hype trying to convince people it can do things it can't (although this a problem with a lot more than AI). For example, this PR https://github.com/mamedev/mame/pull/15031 made me a bit angry because the guy was certain that claude had written a good change when it had misread a datasheet and he didn't try to look at it himself to see if it was correct.

@gloriouscow Shattered wrote a PGC emulation for MAME. Here's an example of running. It's from 1984, way before EGA.

@gloriouscow That and it's instruction set is awful. Not surprising no one used it except Apricot, Altos and Intel themselves. The former probably got discounts.

@gloriouscow Harry looks to be well insulated.

@gloriouscow The electrons need to slow down to make those sharp turns so the newer card is faster.

@gloriouscow This is a bit like the x68000 where the graphics are generated by three different ics and that causes all kinds of weird corner cases because of the way they interact and the state they don't share.

@gloriouscow How many? All of them, of course

@gloriouscow Sure, it's still an invalid opcode after all. I remember reading somewhere that intel sent people to microsoft in the win30 era and asked what they needed to make windows faster and were confused when the response was "make int 6 faster". Microsoft was then using ARPL in v86 mode for vmm syscalls.

@gloriouscow Fun fact: LES EAX,ESP (0xc4 0xc4) is a NTVDM v86 monitor syscall.