Why do I play chess, despite the fact that robots have practically solved it? Why do I make crafts despite the fact that I could purchase stuff of the same quality, if not higher, for the same cost of materials and no cost to my time? Why do I practice piano?

Why do I watch youtube videos about hobby machine shop work despite not having a metal lathe or mill? Why do I watch channels like applied science, or usagi electric, or PBS space time, when I have no interest in becoming a material scientist, working with 50 year old computers, or becoming a theoretical physicist or astronomer?

It's because its personally enjoyable to me! Human beings are naturally curious creatures! We naturally want to share with and learn from others!!

Then you look at the current day with the "Gulf of America" which fails to be primary, common, or local, and just get disappointed.

"aquire" over "acquire" I will readily agree with, that makes a lot of sense. "suficient" and "comunal" however? absolutely not. those doubled consonants are putting in legitimate work in indicating an alternative stress pattern. its suf-FI-cient and com-MU-nal. but without the doubled letters, my intuitive reading becomes su-fi-CI-ent and CO-mu-nal.

There's plenty of words that could stand a respelling in english, but those are terrible examples to call "egregious". doubled consonates have a clear function in the words they appear in.

I'd argue that its format, a dialog between a novice who is yet still sharp (is able to be the one that makes the connections between ideas) and an expert who guides with pointed questions (fills in background context that is needed for the problem at hand), is one of the most effective ways of formatting mathematical communication. Surreal Numbers by Donald Knuth is another great example of the style I'm referring to.

Nope! Just the same old, well beaten, rotting corpse of a horse in the form of a musical theory that entirely exists to prop up schenkerian analysis. The one that that racist, elitist music programs have been teaching since well before the 21st-Century. (The way you can tell is because it includes figured bass, which is totally irrelevant to any discussion that isn't about the harmonic style of 18th century european musicians.)

God forbid I want to, oh I don't know, make music that's similar to the kind that I actually listen to on a regular basis? Better understand its underpinnings and what makes it work in the ways that it does? No. Figured bass.

I think it would be meaningful to also compare the fatality rates of the time tested and proven forms of public transit to these self driving cars too.

Incredibly impressive on a conceptual level though. If someone proposed doing this while having the executable only grab a read-only reference to itself, I think that would be a legitimately solid idea.

Truly, good design is design that goes unnoticed.

My first read was "if you could redo your life from age 17 what would you do". To which "figure out how to make an LLM" would be an insane answer.

I think his advice here is maybe... a year or two too late to be good advice. I can't foresee the job market for machine learning experts being better than it is now in 5-6 years time when said hypothetical 17 year old would be most ready to start career hunting. Either the bubble is gonna pop and the market will be flooded with laid off AI talent.

Even if I'm wrong about there being a bubble at all, I still think that in 5 years time the tech will just have matured to the point of diminishing returns on refining existing architectures. Plateauing until some PHD comes up with something as ground breaking as attention.

Sounds like the idea is, if you've got a bunch of light emitters spaced far closer to each other than the wavelength of light they are emitting, the light output becomes quadratic on the number of emitters. as opposed to linear on the number of emitters in the standard case. And the reverse becomes true as well.

Based on that, I'm pessimistic on the potential of this, as it sounds like charging requires a conversion from electricity to light, and then discharging would also be a conversion from light back to electricity, and we don't exactly have the most lossless processes for that conversion.

Maybe useful in certain kinds of optics or laser work though, where directly dumping the stored energy as light would be desired?

Of course, gut instinct can be wrong, so I checked in godbolt and was surprised to find out that even if I try to force lto, it still doesn't get inlined and remains a call.

And since the bar for validation is expressly stated to be rather weak, I guess this would be best conceptualized as a specialized search engine that can weed out the <80th percentile of mediocre formalized proofs, with a target audience of professional mathematicians who have the ability to make determinations on the final 20% themselves.

Maybe also something I'd wanna skim over at some point, as a non-mathmatician who just thinks lean is neat. Its not useful to me but its fun to learn about.

Oh! If you want that, you should run your own model and set the generation temperature to 0 :) Because that's not what any commercial LLM is doing. Never has been. This is just making up a universe that doesn't exist so you can get mad about no longer being in the universe that doesn't exist.

The masking technique of using a subset of the statistical distribution for each next token isn't going to be meaningfully distinguishable from a natural language perspective. I honestly think its a very elegant way to implement watermarking. I've got no real opinions on how effective it will be to people actively trying to defeat it, but I suspect that the people who are trying to pretend that LLM text was something they wrote themselves are probably too lazy to put in the work to try and defeat it anyway.

Then again, I do mostly remember this in terms of the yet unsolved magic-square-of-squares problem, not the standard magic square.

No space for this human here, I guess.

Edit after reading through this on a seperate device that hasn't been banned from most of cloudflare: I think this guy is vastly underestimating the amount of humans that are still in the production loop of almost every supply chain.

Consider the humble through-hole LED. A product that surely has reached near the absolute peak of production line optimization, being an almost purely fungible product who's form factor will never change. They are still manually put together by humans. A sheet of semiconductor diodes is separated with a manually operated machine for human access. the prongs are placed into a jig by hand, so that a semi-automated machine can place each diode onto the correct leg. Then the jig is placed, again by hand, into position for the injection molded plastic dome.

It's not just one long automated pipeline that has a couple of hoppers for raw materials at one end and finished LEDs at the other. real human beings are still involved at every step of the process. If we can't achieve end-to-end automation for something as dead simple as the humble through-hole LED, after almost 50 years of process improvement? Technology will never fully remove the human from the economy.

Another example of why technical writing is difficult, I think.

I hope the StS team is made aware of this and is able to make the earlier outcomes a bit more evenly spread, so that the distribution matches more closely with what people would intuit them to be.