Unmaking Sense

Episode 11.27: Why AI will force us to rethink what we think matters and our relationship with it.

AI will not just alter our world by taking our jobs: it will force a rethink of the basis of how we regard ourselves, our motives, our values, and what we think matters. A central feature of that re-evaluation will involve the relationship between “is” and “ought”, and it’s coming more quickly than we think.

Episode 11.26: The crises caused by belief in sources of truth and meaning that lie beyond the world

As a further preparation for our discussion of what matters we explore the legacies of religion and metaphysics that insist we defer to sources of truth and meaning that lie “beyond the world”.

Episode 11.25: MoE (Mixture of Experts) approaches to AI are already appearing. What they involve.

The AI wars are hotting up with much current disparagement of OpenAI, competition from Google Deepmind’s Gemini, and now a new version of the Mistral model based on MoE technology called “Mixtral” because it mixes experts. The pace of development is extraordinary and that before the AIs themselves get in on the act.

Episode 11.24: If our sense of what matters is illusory, solving a problem could make it worse.

“Unmaking Sense” means freeing ourselves from the illusory senses we make of the world in order to see more clearly what better sense we could make of it.

Episode 11.23: Google’s Gemini from DeepMind and some of its implications for education and life.

This week Google announced the newest AI tool called “Gemini”. It is an integrated multimodal platform that doesn’t stitch together AIs that were originally trained on separate areas, such as video and audio and text and image. It starts from an integrated base and can therefore deal with all of these things straight out of the box. It can mark homework, correct it, and suggest new things ways to approach it, new ways to process it and it can provide new problems to solve to practise the same skills. The teacher as traditionally conceived is redundant. But what about motivation, about the question why it’s worth anyone ever learning this at all?

Episode 11.22: How Q-learning overcomes the need to search to the end of every branch of a tree.

The problem with chess is that searching beyond a few moves ahead causes a combinatorial explosion: the number of possibilities overwhelms us. So how does Q-learning overcome this problem?

Episode 11.21: Solving mathematical problems using Process (PRM) and Outcome Reward (ORM) Models.

A footnote to the OpenAI saga involving trying to improve the AIs’ mathematical reasoning.

Episode 11.20: Q-learning and AIs that change their own methods of operation.

A continuation of 11.19.

Episode 11.19: Q* = The latest speculation about what really happened at OpenAI regarding Altman.

The twitterati (exorati doesn’t quite do it) are speculating that what the board were concerned about over Altman’s alleged “lack of candour” concerned undisclosed research into an amalgamation of GPT technology with Q-learning that could give (or in some versions already has given) rise to an AI capable of cracking the top encryption algorithm in use: AES 192. Nobody knows for sure, but the speculation is rife.

Episode 11.18: The Dorothea Brooke Syndrome and why nobody knows what is going on most of the time.

Reading the future is computationally expensive, which is why so few can do it at all and most can only imperfectly for very brief periods of time. AGI might be able to do better.