60654

Absolutely. They hooked up an LM and asked it to talk like it's thinking. But LMs like GPT are token predictors, and purely language models. They have no mental model, no intentionality, and no agency. They don't think.

This is pure anthropomorphization. But so it always is with pop sci articles about AI.

> If your premise above is true, then we'd expect to see similar sorts of variations in the various fonts derived from drafting style

If you look at the photos in the article, there are a lot of variations! For example, in the article, if you look at the first two sets of photos of keyboards, you see a variety of shapes, especially visible with the 6s/9s, 0s, Rs, Ss, etc. And then in the next set of photos (the ones with a selection of plaques), you again see a collection of various letter shapes - look at the varying shapes of Gs, Ss, etc. This repeats throughout, when you look at the random assortments of plates and signage.

Later on, after he discusses ANSI and DIN standards, the author goes on to say:

> In the regulatory space, the U.S. military canonized Gorton in 1968 as a standard called MIL-SPEC-33558 for aircraft and other equipment dials, cancelled it in 1998… then brought it back again in 2007.

Except that the specimens he shows right below, of ANSI Y14 and MS 33558 (and whatever the third one is), are very different from Gorton and even from each other - just look at those letter forms. Which makes sense, as their lineage is _not_ from Gorton, but from traditional lettering.

So that's what I mean - it's not that Gorton _specifically_ is everywhere, it's just that draftman's lettering style is everywhere, and in many variants, including the very popular Gorton one.

Of course I read the details on stencils and patterns in the text. But you misunderstand what I'm saying.

What I'm saying is that there are many "concrete instantiations" of drafting lettering style, that all look basically the same, because they all came from the same source, and Gorton is just one of them. So what we're seeing in elevators and on plaques is not "Gorton" specifically. While in contrast we do see Helvetica specifically on NYC subway signs, Johnston in the London tube, etc.

Too fine of a point? Perhaps. And also, it doesn't take away from the quality of the essay which is a delightful romp through the history of draftman's lettering showing up in all sorts of forgotten utilitarian places.

(But I've got to ask - what's with the ad hominem at the end? We should be above that.)

That's an assumption, I read it just fine. :) I just disagree with how the author keeps talking about draftsman's lettering as if it were some company's font (e.g. Gorton and Leroy) rather than a commonly taught community standard.

I did...?

I'm afraid the author has it backwards. This style of writing didn't start as a font from some company.

This is just how you used to be taught to do lettering in a drafting class. The straight lines and simple shapes make it easy to do lettering with pencil or a pen, which is also why all the lines are the same width and all the line caps are rounded.

Later on these turned into stencils, which let users do lettering quickly by tracing the pen inside the stencil - and then turned into fonts for use in print. But all that came later. What we have here is just lettering people learned in drafting classes, and which was later used to make stencils and templates.

I would just add that Mathematica-style term rewriting (e.g. analytical integration, or equation solving) is done with _semantics-preserving_ symbolic solvers, which are hand-made and human-reviewed to guarantee correctness.

LLM style pattern matching and rewriting does not preserve semantics, except accidentally due to an overwhelming amount of examples.

There are people working on that! Google for LLM + PDDL.

There are big problems with hallucinations because LLMs are not smart enough to know when they're starting to make mistakes.

But there's lots of work in this area, and generally in different ways to nail neural and symbolic systems together.

This is exactly right.

Planning and reasoning have long standing meaning in AI, based on explicit knowledge representation and inference. It's a subspace of AI that predates LLMs by decades.

But LLMs don't do that kind of planning and reasoning, and CoT is very clearly not a reasoning technique in the AI sense.

Interesting fact: Unity doesn't use a Microsoft GC, or the Microsoft implementation of CLR.

They had integrated the competing Mono implementation early on, and it came with its own "stop the world" Boehm GC.

They've been trying to move to Microsoft runtime for years but it's slow going.

Yeah, agreed.

CHI, yes, sometimes.

But never ORD - or MDW or GYY. Although CGX would be hilarious. :)

Also, pros: world-class architecture, fun music scene, great museums and of course The Art Institute, top universities and everything that comes with that, great restaurants from every cuisine under the sun, very immigrant-friendly city, lovely tree-lined streets with block parties, I could go on and on. Oh, and living is affordable and people are Midwest Nice.

Cons: it's the great flatness - you have to drive for hours to see hills. Tech industry is ok but very b2b and fintech focused and not a lot of consumer or entertainment stuff. Also as a very liberal city it's a favorite target of culture warrior types.

Tie: winter. Yeah, it gets cold and snowy. But also, it's "just weather". It's the same every year and people deal with it pretty easily.

There appear to be two very interesting results:

1. We can observe how the state machine gets generated, first just a jumble of locations in a hub and spokes topology (no correlations), then some correlations start happening pairwise, making a kind of a beads on a string topology, and then finally the mental model snaps marvelously to two completely separate paths that meet at ends. It's amazing to see these mental models get formed in vivo out of initial unstructured perceptions.

2. In addition to standard HMM modeling, authors find that a "biologically plausible recurrent neural network (RNN) trained using Hebbian learning" can mimic some of this (but not exactly). But more interestingly, they find that LSTMs or transformers cannot. Which makes sense structurally, but it's a good reminder for those who believe the anthropomorphic hype that transformers have memory or other such (they don't :) ).

The scanning is indeed very intrusive, though.