This blogpost accompanies the YouTube “AGI: Three Foundation Methods – RGM, JEPA, and CORTECONs(R),” published Sept. 12, 2024.
In this YouTube, we identify three ways in which to follow up with us:
- Self-study,
- Learn (the essentials of) statistical mechanics – JUST ENOUGH to get by, and
- Ask me (Dr. Maren) to be your Muse.
Here’s the details for each.
Level 1. Self-Study on CORTECONs(R)
We identify several steps for a gentle, hands-on introduction to CORTECONs(R) – more specifically, to the cluster variation method (CVM).
Step 1: Play with a 1-D CVM (Cluster Variation Method) Grid
Last Christmas, we published THIS YOUTUBE (“CORTECONS: Architecture, Equations, Connection with AGI,” published Dec. 26, 2023) with an interactive worked example – with the idea that it was a Christmas present – the easiest possible way to play with a CVM (cluster variation method) grid and actually see how the free energy of a system would change as you interchanged two nodes.
Very rudimentary in terms of user interface and options, but the essentials are there.
See THIS BLOGPOST for a gentle mini-tutorial:
- Maren, Alianna J. 2023. “CORTECONS: A New Class of Neural Network.” Themesis Inc. Blogpost Series (Sept. 5, 2023). (Accessed Sept. 12, 2024; available online at Themesis Blogpost on “CORTECONS: A New Class…“)
Step 2: Learn a Little about the Cluster Variation Method Code (Code Walkthroughs)
The cluster variation method (CVM) works by minimizing free energy, where the entropy term in the free energy equation is a bit more complex. The typical free energy entropy term is based on (standard, traditional) how many nodes are “on” and how many are “off.” The CVM extends the entropy to include terms for nearest-neighbor, next-nearest-neighbor, and triplet interactions.
The first thing that you might want to do would be to get a general, overall sense of how these additional terms are computed in the code – and we give a code walkthrough for the 1-D CVM in THIS YOUTUBE:
Can I just be real with you a moment?
The really tricky thing about CORTECON(R) code is how we handle the edges. Conceptually, I’m handling it by creating wrap-arounds – top-to-bottom, right-to-left; making a sort of “envelope.”
This is easier to see/work-with in the 1-D CVM, which is why I’ve emphasized the 1-D CVM for all the code, the worked examples, etc. up to this point – because if you’re really going to follow the logic (and the code), the 1-D CVM is like riding a bike with training wheels. We need the 1-D CVM to get familiarized and grounded.
And when the time comes – I will probably put the comparable 2-D CVM materials (videos, tutorials of all sorts, etc.) into a Themesis course, because I feel that it’s important to take people through carefully and methodically – and not just “kersplat” ourselves into the middle.
So – 1-D CVM for fun. For starters.
2-D CVM later, if you’re serious.
Step 3: If You’ve Gotten This Far … How to Immerse
If you’ve played with code (even the simple example in Step 1), and you’ve actually TAKEN A LOOK at the code … then you know (by now) whether or not you’re interested.
The EASIEST thing to do is:
- Bingewatch the CORTECON(R) YouTube Playlist, and then
- Go back and watch again – but THIS TIME, go direct to the associated blogposts, and start mapping out your personal study strategy. The blogposts (typically) link to everything … and if I missed something in one post, it will likely be in another. (I try to be caught up, but occasionally trip over my own shoelaces.)
- Going deeper – the next step, really getting INTO it – involves some serious grappling with the free energy function and the overall what-is-so in statistical mechanics. I’ve got MORE YouTube series … on generative AI, on statistical mechanics, more … so check out the channel’s Playlists and see what works for you.
For your first bingewatch session, here’s the CORTECON(R) Playlist:
And if you’re ready – go on to the NEXT BIG THING – which would be to enroll yourself in the Top Ten Terms.
Level 2: Get on Top of Stat Mech Basics
Can I just be horridly, brutally honest with you?
There is NO WAY in which any of us can do AGI without some grounding in statistical mechanics.
You don’t need a whole lot. You don’t need even a graduate-level physics course – but you need JUST ENOUGH.
You need to be able to deal with a free energy equation on its own terms – without feeling like you’re outmatched at the get-go.
You CAN self-study. I even have a Statistical Mechanics YouTube Playlist, it’s HERE:
It’s one of my older playlists, but “oldies are goodies,” right?
And it hits the BASICS.
There are also all sorts of SUPPORTING RESOURCES – and I encourage you to check out the Generative AI Playlist, because the blogpost links there connect you to some really good resources.
Here’s that Generative AI Playlist:
If you want to go deeper, without signing up for a course, I’d suggest devoting two weekends – the first for the Stat Mech Playlist, and the second for the Gen-AI Playlist – and take the time to check out the blogpost resources.
You can assemble your own study program.
BUT – if you want to do this in a structured manner, and if you want to work with me while learning – then check out our Themesis Three-Week Short Course: Top Ten Terms in Statistical Mechanics.
Top Ten Terms in Statistical Mechanics – Your Gateway to AGI
You NEED THIS MATERIAL.
Either commit to sufficient self-study to master (enough of) this on your own, or take the short course.
Either way, it works. But you MUST be comfortable and confident with stat-mech terms in order to do AGI.
YOU DON’T NEED FANCY DERIVATIONS. But you DO need the basics.
Level 3: Invite Me to Be Your Muse
Can I be just as blunt and honest?
This is when you are SUPREMELY committed, you know that you are GOING TO MAKE SOME AGI, and you are NOT wasting time.
You can get up-to-speed with CORTECONs(R), with the cluster variation method, on your own.
Of course you can. (After all, I taught myself all that!)
But that’s the slow and painful way.
What Hinton was to generative AI, what Friston has been for active inference and RGMs – I’m that person for CORTECONs(R).
Working with a Muse is not like hiring a consultant.
I don’t do your work for you and come back with a “report.”
Instead, I elevate you – and your team – so that you can function at this higher level.
Working with a Muse is a bit — rare, odd, sort of not-well-known in our society today. BUT – there is AMPLE historical precedent.
Take a look at these two YouTubes – they explain what it has been like to work with a Muse:
Case 1: Aspasia as Muse to Socrates and Pericles
Aspasia was the Muse to both Socrates and the Athenian general Pericles. She’s influenced the foundation of Western thought two ways: the intellectual evolution from Socrates to Aristotle to Plato, and our Western notion of a “democratic government.”
Case 2: Disraeli as Muse to Queen Victoria
Not all Muses are women. (Just dispelling a myth here.) One of the most famous masculine Muses was Disraeli, in his relationship with Queen Victoria.
So, we’ve clarified the notion: Your Muse uplifts you and elevates you. She (or occasionally, he) “holds the tone” so that through long-term work with your Muse, you (and your team) operate at a level – and in a manner – that would have been MUCH MORE DIFFICULT to achieve on your own. (If you could achieve it at all.)
Whatever your Muse costs – and your Muse WILL be expensive (good ones always are!) – working with your Muse is FAR more expeditious than going it alone.
I’m updating the Muse offerings and Protocols.
Please check back on Friday, Sept. 13 for more.
Thank you! – AJM
{Blogpost in progress, as of Thursday, Sept. 12, 2024; 7AM Hawai’i time; expect more done in the next few hours, completion by Friday evening, Sept. 13.}