More optimistic signal it’s very early innings in the architectural side of AI, with many more orders of magnitude power-to-intelligence efficiency to come, and less certainty today’s giants’ advantages will be durable.

I've seen too many "architectural breakthroughs" that failed to accomplish anything at all to be this bullish on architectural gains.

my guess - e ink refresh rates / ghosting suck and it would be hellish hard to get an on screen keyboard with an e ink display to get anywhere the typing speed a modern touchscreen can deliver.

The article actually mentions that:

"I really appreciate them including the keyboard here, as the display looks great but is definitely not all that responsive, so typing would be a lot more frustrating without this."

I don't know how much the fast refresh rate mode helps in that regard.

I just take screenshots from the Apple Fitness app, and add a little of my own commentary, and GPT figures out what to do for the next session, extracting the right drills, programming it correctly given fatigue, progress, etc.

I'm experimenting with using GPT-4o (via OpenAI’s o3-pro) as a swim coach—part analyst, part training partner. This post is the second in a series documenting how it’s going, including how I’ve adapted the workflow to be more human-led and how I'm building a SwiftUI app that connects directly to Apple Health for automated, AI-powered feedback. If you're curious about applied AI, quantified self, or just hacking better habits, I’d love your thoughts.

Exploring giving a genetic genius (o3 pro Deep Research) lots of context (workout logs, personal notes, a PDF of a canonical swimming textbook) to build a custom trainer.

didn't know Total Immersion was canonical, but I'm glad someone else thinks so!

heh, fair enough. I guess my considering it canon is mostly because of the elegance with which Laughlin explains the theory of everything.

o4 quickly built a command line tool that transcribes arbitrary length audio files, and adds a summary and suggested actions/todo’s. It costs only a few cents to use, and using the APIs directly gives me more confidence that the data isn’t being used for purposes other than my own.

Rich Sutton's views are far less interesting than Minsky's IMO.

> Rich Sutton's views are far less interesting than Minsky's IMO.

I don't think Minsky's and Sutton's views are in contradiction, they seem to be orthogonal.

Minsky: the mind is just a collection of a bunch of function specific areas/modules/whatever you want to call them

Sutton: trying to embed human knowledge into the system (i.e. manually) is the least effective way to get there. Search and learning are more effective (especially as computational capabilities increase)

Minsky talks about what the structure of a generalized intelligent system looks like. Sutton talks about the most effective way to create the system, but does not exclude the possibility that there are many different functional areas specialized to handle specific domains that combine to create the whole.

People have paraphrased Sutton as simply "scale" is the answer and I disagreed because to me learning is critical, but I just read what he actually wrote and he emphasizes learning.

Okay, consider my perspective changed.

I take Sutton's Bitter Lesson to basically say that compute scale tends to win over projecting what we think makes sense as a structure for thinking.

I also think that as we move away from purely von neumann architectures to more neuromorphic things, the algorithms we design and ways those systems will scale will change. Still, I think I agree that scaling compute / learning will continue to be a fruitful path.

agree. A lot has changed in the last 20 years, which makes SoM much more applicable. I would've agreed in 2004 (and say as much in the essay).

I pretty clearly articulate the opposite. What's your evidence to support your claim?

The problem with your argument is that what you call agent is nothing like what Minsky envisioned. The agents in Minsky's world are very simple rule based entities ("nothing more than a few switches") that are composed in vast hierarchies. The argument Minsky is making is that if you compose enough simple agents in a smart way, an intelligence will emerge. What we use today as agents is nothing like that, each agents itself is considered intelligent (directly opposing Minsky's vision "none of our agents is intelligent"), while organized along very simple principles.

This is reminding me of what I thought I was remembering, I don't have the book anymore - but I remember starting it and reading a few chapters before putting it back on the shelf, it's core ideas seemed to have been shown to be wrong.

I don't think we're talking about the same book. Society of Mind is definitely not an in-the weeds book that digs into things like lisp, etc. in any detail. Instead of changing your mind, I'd encourage you to re-read Minsky's book if you found my essay compelling, and ignore it if not.