heyitsguay

LLMs are butchering particular (digital) channels for whimsy and wonder, and it's easy to feel like those channels are everything but they haven't been that way for long and can easily change.

My view: every large online space was basically ruined by an ever-growing advance of inauthentic content pressures starting around 2015. The pools were already pissed in, LLMs are just producing enough piss that, finally, everyone can see it. It's a nuisance in the moment but in the medium term, the human social response will be for the better.

Creating a basic ergonomic example of the full train/deploy loop, using your data engine, for a popular platform in the zeitgeist, will give you a traction boost.

You don't have to innovate on the parts outside your core competency. Have your favorite LLM ingest the Reachy docs, give it control of a robot, and put a lightweight example together. Iterate till you like the example flow and you've purged docs and output text of AI slop, then release!

Yeah similar to how image generators don't draw Mickey Mouse, but ask for a black cartoon mouse with big ears and see what you get!

Products are something other than the sum of their technical parts.

> This gestalt isn't just for biological organisms, but any system for which its decision making engages with representations of the external environment unified with a self-representation to form a coherent representation of a persistent entity engaged with an external world.

This doesn't seem quite right, or at least underspecified. We can talk about this stuff concretely these days, at least in the context of digital systems. E.g. i can draw up a diagram of a system that takes in some camera and audio data (and tactile, proprioceptive, etc.), tokenizes it then runs that + past state data through some autoregressive VLM to drive an inference process. The state being passed around can be written out analytically for a given trained model - the external and internal environmental representations, the linear algebra that transforms them into latent action representations, the process by which that is transformed into control signals. It seems difficult to claim that the computational process that implements this has any more or less of a gestalt then one multiplying two matrices together. So it's not just the existence of certain representations or computational loops that seems to lead to possessing a gestalt.

The two outputs (an image and a string sequence) are learned from entirely different data, even when they're generated with the same model, which is not always the case. There's nothing intrinsic to the learning process that connects image generation and procedural graphics code.

If one wanted to try that, it would likely take the standard LLM finetuning route - scrape a bunch of data pairs at some expense, write some prompts, and have at it.

It was the same sniper that blew up that Starship

What were the results of 3d fractal shader pretraining?

Interesting, can you explain why?

Is it inflammatory if it's true? Seems rather embarrassing (but fitting for 2026 America) to shy away from painful truths because they're uncomfortable.

Moloch!

Thanks for this answer! I appreciate the clarity, I can see the economic impact for your company. Very cool.