AGI Is Still 30 Years Away – Ege Erdil and Tamay Besiroglu

An LLM cannot be placed in a simulated universe, with an internally consistent physics system of which it knows nothing, and go from its initial state to a world-spanning civilization that understands and exploits a significant amount of the physics available to it.

I know that is true because if you place an LLM in such a universe, it's just a gigantic matrix of numbers that doesn't do anything. It's no more or less intelligent than the number 3 I just wrote on a piece of paper.

You can go further than that and provide the LLM with the ability to request sensory input from its universe and it's still not intelligent because it won't do that, it will just be a gigantic matrix of numbers that doesn't do anything.

To make it do anything in that universe you would have to provide it with intrinsic motivations and a continuous run loop, but that's not really enough because it's still a static system.

To really bootstrap it into intelligence you'd need to have it start with a very basic set of motivations that it's allowed to modify, and show that it can take that starting condition and grow beyond them.

You will almost immediately run into the problem that LLMs can't learn beyond their context window, because they're not intelligent. Every time they run a "thought" they have to be reminded of every piece of information they previously read/wrote since their training data was fixed in a matrix.

I don't mean to downplay the incredible human achievement of reaching a point in computing where we can take the sum total of human knowledge and process it into a set of probabilities that can regurgitate the most likely response to a given input, but it's not intelligence. Us going from flint tools to semiconductors, vaccines and spaceships, is intelligence. The current architectures of LLMs are fundamentally incapable of that sort of thing. They're a useful substitute for intelligence in a growing number of situations, but they don't fundamentally solve problems, they just produce whatever their matrix determines is the most probable response to a given input.

It is science fiction to think that a plane could act at all like a bird. Although... it doesn't need to in order to fly

Intelligence doesn't mean we need to recreate the brain in a computer system. Sentience, maybe. General intelligence no

Let's say that the shortest interval at which our hearing has good frequency acuity (say, as good as it can be) is 1 second.

In this interval, we have 44100 samples.

Let's imagine the samples graphically: a "44K" pixel wide image.

We have some waveform across this image. What is the smallest frequency stretch or shrink that will change the image? Note: not necessarily be audible, but just change the pixels.

If we grab one endpoint of the waveform and move it by less than half a pixel, there is no difference, right? We have to stretch it by a whole pixel.

Let's assume that some people (perhaps most) can hear that difference. It might not be true, but it's the weakest assumption.

That's a 0.0023 percent difference!

One cent (1/100th of a semitone) is a 0.058% difference: so the difference we are considering is 25 X smaller.

I really don't think you can hear 1/25 of a cent difference in pitch, over interval of one second, or even longer.

Over shorter time scales less than a second, the resolution in our perception of pitch gets worse.

E.g. when a violinist is playing a really fast run, you don't notice it if the notes have intonation that is off. The longer "landing" notes in the solo have to be good.

When bad pitch is slight, we need not only longer notes, but to hear it together with other notes, because the beats between them are an important clue (and in fact the artifact we will find most objectionable).

Pre digital technology will not have frequency resolution which is that good. I don't think you can get tape to move at a speed that stays within 0.0023 percent of a set target. In consumer tape equipment, you can hear audible "wow" and "flutter" as the tape speed oscillates. When the frequency of a periodic signal wobbles, you get new signals in there: side bands.

I don't think that there is any perceptible aspect of sound that is not captured in the ordinary consumer sample rates and sample resolutions. I suspect 48 kHz and 24 bits is way past diminishing returns.

I'm curious what it is that Deadmau5 thinks he discovered, and under what test conditions.

If you feed that true randomness into a computer, what use is it? Will it impair the computer at the very tasks in which it excels?

> That all inputs should be defined and mapped to some output and that this process is predictable and reproducible.