Do you posit that there are enough examples of 30x8 ellipses encoded in braille online for ChatGPT to learn from but not 31x7 or 31x9 ellipses? That seems unlikely.
Yes, or the model got lucky with the quality of output for a particular combination of my prompt and the reasoning behind its answer that lined up with something it had seen before — quality which it was unable to recreate under slightly different circumstances.
They lost me a little before then - Claude Code's regressions were so very obvious and there's no sign they've learned their lesson in this article or in the comments of those who work on Claude Code on HN. They'll continue to tweak and generally mess around with a product people are using, altering the behaviour without notice in ways that can severely impact use, for months! GPT5.4 has been remarkably consistent and capable, as a replacement. I've cancelled my max plan.
In the video in question, he doesn't seem able to choose a good scoring function for the stochastic solver (even over multiple weeks), seemingly choosing a linear sum of distances (see 8:50) between simulation and reality. That's a mistake that not even an undergraduate should make. He needs some domain experts.
Yes, that's another great example of the same kind of thing - creating a JIT from an interpreter. It remains true that interpreters do not directly generate machine code.
What are your goals, to let everyone know that interpreters, definitionally don't generate code? This isn't debate club.
I dropped a cool link that shows we have a machine that turns interpreters into compilers. I am talking about the machine. You are talking about the definition. We aren't talking about the same thing.
Partly, it's simply that words matter. An interpreter is not a compiler, even if partial evaluators and Futamura transforms are very cool. Posting about them in a context that isn't a confusion about what interpreters are may have been more fruitful.
This is still incorrect. A bytecode interpreter, as its name indicates, interprets a bytecode. Typically, compiling a bytecode to native machine code is the work of a JIT compiler.
True, but that’s using “effects” in a broader sense than people seem to mean here. The discussion seems to be about the visible effects the audience experiences as effects, and whether those age well, not invisible digital cleanup, compositing, or set extension.
It's not true either way. Very little is actually practical, it's just that when something looks good people think it is practical because they want to believe that.
Marketing feeds into this and tells people movies were done all practical or made "heavy use of practical effects" and it's just lies.
Even before this people were saying stuff like mad max was done "almost all practical" because they saw behind the scenes stuff of flipping a few cars even though the movie is wall to wall digital effects. Sometimes the elaborate "practical effects" don't even move right and are used for reference and completely replaced.
reply