Remix.run Logo
▲ lossolo2 a day ago

Aren't we approaching the limits of physics, with increasingly diminishing returns each generation? I remember reading an article about quantum effects at these scales, which make everything much harder. I mean, for example, even if we double the speed at which we can travel every year, you will never get to Alpha Centauri in your lifetime

▲the8472 a day ago | parent | next [-]

Chips are only 2D, and we're far from the landauer limit, and the landauer limit only applies to non-reversible computations anyway. If we could assemble matter atom by atom at scale we could get to some dirty diamond that fits in your hand, outclasses the world's top supercomputers and runs off 10W of sunlight.

▲thombat a day ago | parent | prev | next [-]

The overall point is sound, but in fact if we could double our speed every year then in just 23 years we'd get from 100km/h to the speed of light

▲ignoramous a day ago | parent [-]

Keeping with the Interstellar framing, the key is not only in faster ships (lightspeed) but in shrinking distance (warping space-time). For LLMs, that'd mean efficient model architectures on efficient chips.

▲GoblinSlayer 8 hours ago | parent | prev | next [-]

That's the Electron factor. Unix worked on PDP-7 with 18kb of RAM.

▲dist-epoch a day ago | parent | prev [-]

We are approaching the limits in a single chip. But we can add more chips, millions of them, billions of them. Gigawatts of compute.

▲twoodfin a day ago | parent [-]

Right. I’m sure at some point in the past more capital was spent on designing better airplanes than on building them & the infrastructure to support them.

We still spend money on designing better airplanes, but the two lines on that graph crossed long ago.