The Test of Prometheus

I feel like the argument here for &quot;not passing the Turing test&quot; is simply a reframing of the test.<p>Which is fine. But a better conclusion is &quot;chatGPT passed the Turing test, but all that means was that the Turing test was insufficient to determine intelligence.<p>Once upon a time we might have defined intelligence as the ability to do math really really fast. Or play chess really well.<p>Clearly those tests are insufficient. Equally the Turing test has been shown to be insufficient.<p>I look forward to the next-generation intelligence test. And yes, maybe the Prometheus test is it.

Now that was a great read.<p>The idea where the difference is power vs empowerment.<p>This makes AGI feel closer. Also that it all starts with a heartbeat. It defines time, it narrows purposes and gives will.

This article makes the same mistake that Turing made: assuming that by creating a thinking machine one is creating a mind. We are not gods, we have simply made a machine. We haven’t externalized intelligence either. Intelligence isn’t determined by our thoughts alone, but through our actions, our will.<p>I critique many of these false assumptions in an article of mine:<p><a href="https:&#x2F;&#x2F;leroy.works&#x2F;articles&#x2F;a-critique-of-alan-turings-conclusions-on-thinking-machines&#x2F;" rel="nofollow">https:&#x2F;&#x2F;leroy.works&#x2F;articles&#x2F;a-critique-of-alan-turings-conc...</a>

Meh. Adding a &quot;free will&quot; is not an interesting problem. Paperclip maximizer has a free will that is rather trivial to program over GPT-4 API. It always was the intelligence but that is hard.

<p><pre><code> Put a lone human in one closed, isolated room, and a machine in a different one. You only have a text-chat interface to each of them (as per Turing&#x27;s original Imitation Game set-up). Start a conversation by saying hello and asking a random question. After receiving a satisfactory answer (which you can ignore), simply wait for an hour, 12 hours, 48 hours, 1 month. </code></pre> I want to ask a similar but simpler question.<p>» Put a lone human in one closed, isolated room, and another lone human in a different one. You only have a text-chat interface to each of them (as per Turing&#x27;s original Imitation Game set-up).<p>Now would people when given the question, which one is a person and which one is a machine be able ignore the question asked and tell they are both humans?

As I mentioned elsewhere in this thread, when talking to state-of-the-art LLMs like GPT4, we should remember that they&#x27;ve been fine-tuned to be servile and give answers that are likely to please humans (and with a big helping on censorship on top).<p>If we are serious about testing their sentience, we should really be talking to the raw models.

I thought that this true AI thing would be something for twenty years from now, but this is getting pretty concrete.<p>Promising and scary.

I suppose I shouldn&#x27;t be suprised that the goalposts for &quot;real AI&quot; are now &quot;ah but does it need to piss?&quot; :P