https://www.youtube.com/watch?v=02pBnDkV0rQ

Einstein is not hyped enough

Einstein discovered atoms, derived the lorentz transform from the principle of relativity, laid a foundation for the quantum hypothesis, created a theory of gravity that outdid Newton, and on top of it invented the statistical interpretation of quantum mechanics. The guy was a genius

Also, at this point it's really only maybe 1 order of magnitude more storage than he imagined. The model that passed the test was GPT-4.5. There are now open weight models like Gemma 4 and Qwen 3.6 which you can run on your own computer if you have a graphics card with 32 GiB of RAM (or even 16 GiB of RAM, but you have to quantize it enough that you lose a significant amount of performance), which perform better than GPT-4.5 in most benchmarks.

Now, I don't know if anyone has run a full Imitation Game with them, performance by LLMs can be quite spiky so they can be good on some benchmarks but bad at other tasks. But in general, these ~30B parameter models that you can run locally now outperform GPT-4.5 on many common tasks, so it's looking like he was only really off by about 1 order of magnitude, and a quarter of a century.

It describes the problem and objections quite well. For instance, I believe that "May not machines carry out something which ought to be described as thinking but which is very different from what a man does?" is absolutely true of current LLM chatbots.

This also appears to be true of LLMs: "We also wish to allow the possibility than an engineer or team of engineers may construct a machine which works, but whose manner of operation cannot be satisfactorily described by its constructors because they have applied a method which is largely experimental"

We don't, in fact, know exactly how LLMs work, because they are simply enormous neural networks trained via gradient descent. There is a whole field of mechanistic intepretability, of studying how LLMs do particular processes.

"It is probably wise to include a random element in a learning machine. A random element is rather useful when we are searching for a solution of some problem."

Our current LLMs absolutely do use random elements in their learning, and inference, processes.

Finally, a study has been done with a full 3 party Turing Test, as described in Turing's imitation game. And GPT-4.5 with a prompt providing a persona, along with a delay to account for typing speed, has passed it on two different groups of subjects (undergrads, and people hired via an agency): https://arxiv.org/pdf/2503.23674

While what LLMs do is not quite like how humans think, and I wouldn't describe it as consciousness, I think there's a convincing argument to be made that they do think, according to the criteria of Turing's Imitation Game.

Yeah, it took a few order of magnitude more storage, and a lot more speed, than he was imagining. But otherwise, the LLMs of today behave a lot like he imagined; they are trained rather than programmed, they use random elements, they definitely work differently than how humans think.

There are many reasons that LLMs are like models of alien intelligence because they are not like us. But they are more like us than Eliza with a huge database. Lol.

some of the younger queer people on here have come up with the slogan "make sure to be extremely gay on the computer or Alan Turing died for nothing".

it's not, like... we have some professional/activist experience in designing slogans and that's not one we'd have picked, there are many problems with it, but it sure does speak to an emotional truth.