Well, I’ve watched it.
There are a few things to realize.
-
The code it generated ran without an issue with the simulation task. It, however, is incomprehensible without understanding of all the details (like any other code).
This is, probably, so because they have feed in a lot of very similar internal code in the training phase.
The gibberish from the “thinking” phase might be helpful or it may be equally cryptic.
Keep in mind that the connection between “thinking” and the code is illusory (only apparent) – the probabilistic process of inference generates the code from very different area (of the data structure) than the “thinking gibberish” has been generated from. This is just how such models work – they merely traverse a huge tree-like data-structure.
The principle still holds – programming is understanding, not coding. And the crappy code is cheap, indeed.
-
The tetris game seem to be subtle broken, and this is the whole point! – it eventually failed to remove a completely filled row at the bottom. Have good luck to guess why, lmao (code does not come from your own understanding). So you basically have to debug someone-else’s unfamiliar code.
This is OK for a meme-python script. But imagine the degens would generate anything “serious”, and, god forbid, something that controls anything real (and the degens definitely will).
So, again, programming is understanding and it cannot be “outsourced” to some word2vec-based crappy model, however large.
The agents stuff is kind of cool, but it is an imperative code that runs alongside the models, simply because an inference algorithm just selects paths based on estimated probabilities.
And marketing is really cool – become part of the new tech “elite” on X for just so many dollars per month.
This is what Elon is good at – creating and selling the new (emerging) status symbols (which all started, if you think about it, from that Job’s “I am a PC, I am a Mac” absolutely wonderful ad).
And yes, meme-robots will be a nothing but a status symbol too.