[latexr]

That’s a well-remembered video, but I don’t really think it fits with the original comment. What we’re seeing is more akin to already having the plane flying (badly, still crashing frequently and landing in the wrong country) and instead of making it more reliable, everyone is trying different wheels and paint colours.

We’re not really seeing any significant development with this. What LLMs need most desperately (and are far from getting) is reliability and not being convincing liars. Being able to query existing server models from your oven timer is a cool gimmick but not really transformative or advancing anything.

It’s like a reverse JFK Space Effort speech: “We choose to indiscriminately throw at the wall every single LLM-adjacent idea we can think of. Not because it is useful, but because it is easy and potentially profitable”.