Unmaking Sense

Episode 8.71: The addition of plug-ins to the GPT family is understandable, but comes at a price.

Providing access to third-party APIs involves a major new step in the generation of artificial general intelligence. But precisely because all the GPT family so far did not reach out to 3rd parties, to the Internet or other vendors the product of their neural nets arose from what they “knew”. When they can “look up” things external to their own neural net, they cannot “think with them” and so this welcome new stage on the road to AGI is only a stepping-stone: to complete the journey all this will need to be brought inside.

Episode 8.70: How new chatbot technology breaks the grip of the world.

More on the theme of the global changes that will ensue from this chatbot technology. Pastimes and “passtimes” suddenly became all the more necessary.

Episode 8.69: Chatbot coders need to learn the adage “When in a hole, stop digging!”

Chatbots including GPT-3.5-turbo do get things wrong and they do get coding wrong. In particular, if you ask them to help out with your coding, they will very often go forward when the sensible move is to go back. Retracing their steps going back to a previous working state isn’t something they’re very good at; that’s partly because of the way they’re built. If they were to learn that and to give advice to that effect, things would fundamentally change, and I’m sure one day soon they will.

Episode 8.68: The lack of blame in a chatbot, eliminating bullying, promises to change everything.

The fear of making a fool of oneself is one of the principal obstacles and inhibitors to asking the questions that we need to ask. Throughout life, there are many questions that we would like to ask, but do not because we fear that others may laugh at us. But the emergence of a benign universal tutor in a chatbot promises to change all that, and so the world.

Episode 8.67: This is the way the world ends; this is the way …: not with a bang, with a Whisper.

I report back on my experiences with transcribing the digital recordings of the episodes of the podcast and reflect on some of the implications for civil liberties for oppression, and for all of us if this becomes widespread as it certainly will.

Episode 8.66: Electricity and the consumption of power: an exercise in machine learning.

Fresh from the triumph of managing to get through, and actually speak to somebody from my electricity provider. I started thinking about the consumption of electricity by different appliances and particularly by those that sit quietly in the background consuming electricity without anyone really noticing what are called phantom appliances. And I then thought this could be a good test case for machine learning, so I started to find out if I could do it.

Episode 8.65: The arrival of chatGPT-4 and some reflections about coding.

We discuss GPT-4 and consider some data structures in python.

Episode 8.64: The trans-human and it’s implications for human knowledge.

We like to think that the way human beings think is the way the world works. The ultimate human theory is something that can be printed in a formula on a T-shirt. But trans-human artificial intelligence suggests something different: that the universe may not be intelligible at all in terms of such simple theories; that big data may exhibit patterns that rely on more dimensions of simultaneous perception than any human is capable of perceiving.

Episode 8.63: Thinking many things at once in many dimensions. Why chatGPT is already trans-human.

We think about one sense in which a chatbot may already be trans-human. That is because it is trained to deal with multiple dimensions of information both in terms of the training data, and in terms of its neural net architecture, it is intrinsically multidimensional. Human brains at least in their non-conscious form may similarly be multidimensional, but a conscious mind is very one-dimensional and that makes it very difficult for us to understand some of the solutions that a chatbot my present to us. We consider one response by my fine-tuned model based upon Marcus Aurelius, as an example of the dilemma that we face. We also consider the black box problem of not being able to understand, as human beings, conclusions and modes of argumentation that are trans-human because they lie by definition in a domain more sophisticated than we are.

Episode 8.62: Polynomial and Non-Polynomial time. Implications for conversation lengths.

We consider the way a chatbot evaluates the previous conversation, how that makes its computational time increase exponentially and must impose a serious limit on how long any conversation with it can be. We look at that in the context more generally of the way we converse and the chatbot responds to conversations: with humility, gracefulness, affirmation and courtesy.